Multi-senseor satellite precipitation estimates on the African continent using combined morphing and histogram-matching techniques Malte Diederich

1

, Aynur Bozoglu

2

, Clemens Simmer

1

, Alessandro Bataglia

1 1 University Bonn 2 EUMETSAT Collaboration between IMPETUS, AMPE, and Precip-AMMA projects

Multisensor Satellite Precipitation Estimates Goals of presented work:

    Provide high resolution precipitation estimates (Meteosat-7/MSG resolution) in West Africa (IMPETUS Project) Create a product that can be merged with and used to dis/reagdregate ground observations Test and recommend possibilities for upgrades to the EUMETSAT MPE => AMPE Assess product useability for climatology, hydrology, agriculture (as far as possible)

Procedure:

Merge information from geostationary IR images and passive microwave platforms  Probability Matching   Morphing Identify non-raining clouds with SEVIRI – Channels 7 to 10 Examine possibilities for regionalized calibration  Ground validation with interpolated kriged products as well as point measurements

Sensors

Low-earth orbit Satellites Passive Microwave Sensors: •

TMI

(TRMM Tropical Rainfall Measuring Mission) 2A12 Product •

AMSR-E

(AQUA Satellite) •

3 SSM/I

(DMSP F13, F14, F15) •

(3 AMSU-B)

(NOAA 15, 16, 17) (Nesdis or simple scattering Algorithm) Geostationary Satellites

METEOSAT 7:

•30 minutes sampling interval •5 km resolution at sub-satellite point •11 µm IR channel

MSG-1:

•15 minutes sampling interval •3 km resolution at sub-satellite point •10 µm IR channel •Cloud Analysis with some potential for discriminating non-raining clouds

Probability matching of IR and RR:

Histograms Look-up table IR Radiance 1.

2.

3.

IR Radiance Accumulate co-located passive micro-wave rain estimates and brightness temperature measurements:

200 km, +-4 day accumulation windows, spaced at 10 km and 1 day

Match the cumulative distribution function of rain estimates and IR 11 μm brightness temperatures to obtain a look-up table, associating the coldest cloud temperature with the highest rain rate . (histogram-matching) The resulting look-up table translates IR radiances into rain rates Rain Rate

Multisensor Satellite Precipitation Estimate Morphing of PMW rain estimates

Calculation of advection vectors by cross-correlating subsequent IR images (10 μm): Size of windows for cross correlation determines if tops of small clouds or storm systems are tracked.

2 h before propagated backward Scan during Overpass 2 h after propagated forward

Combining Morphing and Histogram Matching

Forward and backward propagated PMW scans propability matching estimates are using a weighted merged with averaging system based on time to the last/next PMW scan: Propagated PMW weights decrease linearly with time from overpass, weight reaches 0 at 2 hours from original scan Example of weighting on one day at 10 degree latitude PMW scan Forward propagated PMW Backward propagated PMW Histogram-Matching Temporal sampling of microwave-overpasses changes with latitude. At 50 Latitiude, histogram-matching component gives negligable contribution

Ground Validation

1. Evaluation of Morphing and Probability matching performance.

2. Regional and temporal distribution of systematic errors Data Sets available for validation:

Benin:

• • • DMN / CATCH / AMMA / IMPETUS monthly accumulations kriged at 0.1

° resolution from a dense gauge network, Daily point measurements from 2002 to 2005 Sahel region: • AGRHYMET procucts for June-September 2004, AMMA intercomparison excercise 10-day accumulations from filtered synop stations 10-day accumulations from 800 stations kriged at 0.5

° resolution

:

• •

Furthermore:

Daily synop data from GPCC for AMPE validation: High density and quality in Europe, scarce and with gaps in Africa Daily Nigerian and South African gauge observations

Ground Validation Benin

Monthly sums for the Benin, June-September 2002 Comparisons with a 0.1x0.1 degree Kriging product (IMPETUS) Histogram Matching with morphing

Regionaly dependant biases Gradients in air humidity and moisture advection from north to south may lead to altered relation between ice in cloud and surface rain

Shape of probability density function of daily estimates agreed well with grouund observations, but skill for correct daily prediction diminishes with intensity 01 mm 11 mm 21 mm 31 mm 41 mm 51 mm 61 mm

False alarm ratio Probablity of detection Skill score

Data quality of rainfall measurements in Benin

Quality of ground measurements can be estimated from time series of satellite estimates. Some stations display other accumulation time than 6 UTC

Reliability of satellite and gauge cross-validation

Satellite/gauge skill at gauge point Gauge/gauge skill as function of distance In addition: Flag single days if ground observation is extremely unrealistic Following validation of gauges in Benin with satellite data: •7% of rain days given by gauges can not have ocured on the associated date •At least 6% of non-raining days (between 6% and 20%) given by gauges should have been rainy days

Ground Validation

histogram matching only with morphing

Lower performance in Europe, but morphing still improves estimates 7.2004

Morphed + IRS morphed Histo.

8.2004

Corr.

0.51

0.49

0.35

Morphed + IRS morphed Histo.

9.2004

Corr.

0.57

0.56

0.44

Morphed + IRS morphed Histo.

Corr.

0.44

0.42

0.28

Bias -18 -14 -24 Bias -26 -23 -36 Bias -38 -36 -44 rmsd 43 43 50 rmsd 52 51 63 rmsd 55 51 60 Lower performance in european test area due to: •underestimation of costal precipitation in microwave products •Orographic rainfall partially recognized by PMW displaced by merging scheme •Very small convective cells not detected by coarse resolution PMW

Infra-red screaning

: Post-Processing of rain product where semitransparent clouds (SEVIRI Channels 8 and 10) and clear sky (Channel 10) are declared to be no rain areas

Conclusions Morphing technique superior to probablility matching It is recommended to recalibrate PMW estimates regionaly, especially coastal areas: •Positive bias in Sahel •Negative bias in Europe •strong bias gradient from moist to dry, especially near coasts Satellite estiamtes are relatively good at detecting rainfall even at point scale, but quantitative skill not so good Even in a relatively dense network like Benin, there are some regions where satellite is better for detecting rain than interpolated gauge products Adding IR screening or classification may inprove morphing estimates Future plans •Validate other regions of Africa with filtered synop stations (GPCC-input, Nigeria, South Africa) •homogeneization of the Passive Micro-Wave estimates with respect to pdf and bias •Test other AMSU-B products •Optimize weighting system beteween morphing and histogram matching with TRMM radar ground measurements •Improve morphing estinates using EUMETSAT cloud type products

Reminder

Vertical reflectivity profile measured by MRR in Benin