satellite rainfall estimation and validation in the frame of AMMA the African Monsoon Multidisciplin

1
satellite rainfall estimation and validation in
the frame of AMMA (the African Monsoon
Multidisciplinary Analysis)

• Michel Desbois, Franck Chopin, Isabelle
  Jobard, Abdou Ali, Abou Amani, Thierry
  Lebel, and the EU Precipamma Group

LMD-IPSL-CNRS Agrhymet Center, Niamey, Niger
LTHE-IRD, Grenoble
2
Special observations and developments for AMMA
AMMA observations are covering different time
periods Long term (10 years and more), Extended
(3 years), Special Observing Periods (in 2006).

• Observations include
• operational networks (meteorology, hydrology,
  aerosols),
• enhanced networks (radiosoundings)
• Specific measurements in supersites
• Aircrafts and ships observations during SOPs
• Specific collection and processing of satellite
  observations (AMMASAT)

Specific databases are constructed to collect and
distribute the data to the project partners,
without operational objectives.
3
Specific needs of AMMA multidisciplinary
Requirements of precipitation products from
different communities Hydrologists, Surface
water budget analysts and modelists
(SVATs), Agronomists, Climate modelists, Mesoscale
modelists, Intraseasonal variability
analysts, Various impacts people, for example
health impacts specialists Lead to requirements
for time and space scales ranging from 10 days, 1
X 1, to hours, 10 x 10 km (continuously). Not
accounting for instantaneous estimates needed for
assimilation in forecast models
4
Requirements for satellite precipitation
estimations during AMMA
Use the best time resolution available in that
area 15 minutes with Meteosat Second
Generation, Ensure the capablity of this
satellite to detect properly rainfall
areas, Ensure the consistency of rainfall
measured at different time-space scales Ensure
the consistency with a reference product at large
space-time scales. Provide estimations of errors
at the different scales retrieved
5
Development and validation of specific
algorithms (precipamma group)

• European group (LMD/CNRS Paris, CNR Bologna
  Ibimet, Univ. Of Bonn, Tamsat Univ. Reading)
• Tests on year 2004 (validation data provided by
  AGRHYMET and IRD)
• First results show satisfactory results of the
  LMD method based on MSG (EPSAT-SG)
• real time application of EPSAT-SG trained on
  previous years on rainy season 2006 (AMMA SOP)
• Methods will be re-runned with complete data
  sets and validated against AMMA-SOP data
  (including radar data). New intercomparison
  exercises planned.

6
EPSAT-SG scheme(Estimation des Pluies par
SATellite Seconde Génération)
Neural Network
MSG IR Channels SRTM Digital Elevation Model
2A25 TRMM precipitation Radar data
Equation 1
Rainfall probability images (Pr)
GPCP1dd rainfall images (Igpcp)
A is a disc of about 125kms radius cA is the
centre of A d is the considered day T is the
period d-15days,d15days dt1 corresponds to 1
day dt2 corresponds to 15 minute da corresponds
to 1 MSG pixel
Equation 1
Potential rainfall intensity images (Ip)
Equation 2
Equation 2
Estimated Rainfall Intensity at time t during day
d and position a
Final product resolutions Space resolution 3
kms Time resolution 15 minutes
EPSAT-SG rainfall estimation (Ie)
7
Collocation between Pr image (2A25 TRMM data -
red- ) and EPSAT-SG probability of rainfall
-gray levels-
8
Example of validation on 10 days periods, 1
degree x 1 degree

• Validation from krieged data
• Provided by IRD and AGRHYMET
• Space resolutions 0.5, 1 and 2.5 degrees
• Time resolution 10-day periods
• The validation datasets have been provided with
  an estimate of its uncertainty for each grid
  cell.
• More detailed data sets inside boxes Niger,
  Benin.

9
Validation (contd)
1x1 rainfall accumulation from the IRD
AGRHYMET Raingauge Dataset during the third
decade of August 2004
10
Validation (contd)
1x1 rainfall accumulation from the IRD
AGRHYMET Raingauge Dataset during the third
decade of August 2004
11
August 2004 Third Decade
GPCP/krieged surface data
EPSAT-SG/krieged surface data
12
Comparison between GPCP 1dd and EPSAT-SG

• Two validation studies have been done
• First one considers all the validation grid
  cells.
• Second one takes into account only the 50
  validation grid cells with the lowest krigging
  uncertainty.

13
Smaller time scalesEPSAT-SG and validation data
for the square degrees of Niger and Benin, for
different time resolutions
rain estimation per event over the degree square
of Niamey for 2004
Probability of rainfall, first decade August
2004, with surface measured rainfall of the full
event
R2 of the time series, for different accumulation
times
14
Application to 2006 Example of near real time
estimation of rainfall from EPSAT-SG for the
needs of the AMMA campaigns
Accumulated over a 3 hours period
15
Application to 2006 Examples of decadal
estimates of rainfall, preliminary operational
version of EPSAT-SG
60 mm
80 mm
gt150 mm
gt250 mm
16
Application to 2006 Examples of decadal
estimates of rainfall, CPC product
lt100 mm
80 mm
200 mm
gt250 mm
17
Application to 2006 Examples of decadal
estimates of rainfall, EPSAT-SG provisional
product
1-10 July
1-10 August
11-20 July
11-20 August
21-31 August
21-31 July
18
Latitude-time Hovmöller of the 2006 African
Monsoon, at two different longitudes (2E Niamey,
7.5W Bamako)
19
Conclusion and future developments

• Epsat-SG gives results compatible with
  operational precip algorithms (CPC, GPCP)
• Quality of the results has been estimated at
  different space-time scales
• It allows users to integrate at the space-time
  scales of interest for them
• Further validations and intercomparisons will
  take place with the full AMMA surface validation
  network operating in 2006
• The method remains basically an MSG IR algorithm.
  Although using a set of channels, it cannot
  detect very high rainfall rates on short time
  periods. The duration and extension of events is
  overestimated, while the max intensities are
  underestimated.
• The method is presently educated by the TRMM
  radar, and tuned to GPCP monthly accumulations.
  Other data sets may replace these entries
  (passive microwave, surface networks)
• The method is transferable for practical
  applications. This is of particular interest for
  African institutions. Tests have to be performed
  over other regions covered by MSG or equivalent
  satellites.
• The education through a space radar is efficient.
  A new precip radar in tropical orbit would be
  useful after TRMM

20
The 13 MSG neural networks inputs

• IR Temperature indicator
• 10.8 µm
• IR multichannel indicators
• 10.8 µm - 6.2 µm
• 10.8 µm - 7.3 µm
• 10.8 µm - 8.7 µm
• 10.8 µm - 9.7 µm
• 10.8 µm - 12.0 µm
• 10.8 µm - 13.4 µm
• Temporal difference indicator
• 10.8 µm - previous 10.8 µm

• Local variance indicators
• Variance 5x5 6.2 µm
• Maximum 5x5 6.2 µm
• Variance 5x5 10.8 µm
• Maximum 5x5 10.8 µm
• Geographic indicators
• Altitude derived from SRTM data