DATA ASSIMILATION OF VOLCANIC ASH RETRIEVALS USING THE NEW GENERATION OF GEOSTATIONARY SATELLITE SENSORS - CHEESE COE
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
www.bsc.es
Data assimilation of volcanic ash retrievals using the new
generation of geostationary satellite sensors
Andrew Prata (1), Leonardo Mingari (1) and Arnau Folch(1)
(1) Barcelona Supercomputing Center (BSC), Barcelona
IUGG 2019 General Assembly, Session JAV02
Montreal, 8-18 July 2019
Center of Excellence for Exascale in Solid Earth
Space-based remote sensing of volcanic clouds
• Satellites are the major source of information for fine ash cloud components (i.e. “away” from the source)
• Circumpolar satellite observations have obvious limitations in terms of data frequency/coverage
2
New generation of geostationary satellites
Satellite Sensor Coverage Spatial res. Temporal res. Ash/SO2 bands Lifetime
(μm)
Meteosat-11 SEVIRI Europe/Africa 3 km 15 min 7.35, 8.7, 10.8 & 2015-2022
12
FY-4A AGRI S Asia/Oceania 4 km 15 min 8.5, 10.7 & 12 2016-2021
Himawari-8 AHI S Asia/Oceania 2 km 10 min 7.35, 8.6, 10.45, 2014-2029
11.2 & 12.35
GOES-17 ABI W America 2 km 10 min 7.4, 8.5, 10.3, 2018-2029
11.2 & 12.3
GOES-16 ABI E America 2 km 10 min 7.4, 8.5, 10.3, 2016-2027
11.2 & 12.3
Meteosat-11 FY-4A Himawari-8 GOES-17 GOES-16
3
Volcanic ash retrieval process
Set of effective Set of optical Satellite data
radii (re) depths (τ)
Ash
1. Pre-compute LUTs detection 3. Retrieved physical
2. Detect ash and
search LUTs parameters
Single-scattering
properties
Radiative Look-up
Refractive Mie
Qext, SSA transfer tables LUT search ml, re,t
index data program
code (LUTs)
Errors in physical (retrieved)
parameters generally estimated to
be 50%
Ts and Tc
Major sources of uncertainty:
1. On microphysical model and LUTs 2. On volcanic ash detection itself
a. Composition (e.g. andesite, rhyolite, basalt) a. Setting the BTD threshold is a balance between false
b. Size distribution (e.g. uniform, lognormal) positives and true negatives
c. Shape (spherical as Mie calculations are exact) b. Underlying meteorological cloud affects optical depth in
d. Cloud surface (Ts) and cloud-top temperatures (Tc) ash-detected pixels 4
FALL3D selected as a European flagship code in ChEESE
Center of Excellence for Exascale in Solid Earth
www.cheese-coe.eu @ChEESECoE
• ChEESE is a Center of Excellence (CoE) for Solid Earth hazards and High Performance Computing (HPC)
• 10 open source European codes will be prepared for the upcoming Exascale machines (2020-2022)
o 2 codes in volcanology (FALL3D and ASHEE)
o FALL3D-8.0 solves atmospheric transport of particles and aerosols
• 12 Pilot Demonstrators (PDs) and enabling of services oriented to society:
o FALL3D-8.0 is on ChEESE PD#12: high-resolution volcanic ash dispersal forecast using ensembles
o Ensemble ash cloud modelling (beyond embarrassing parallelism)
• ChEESE will integrate codes/workflows in EPOS and EUDAT
5
ChEESE PD#12: high-resolution volcanic ash dispersal
forecast using ensembles
• PD#12 will increase resolution of present operational ash dispersal model setups by one order of magnitude
• The pilot will develop an ensemble-based data assimilation system combining an Ensemble Transform
Kalman Filter (ETKF) and the FALL3D ash dispersal model in order to produce:
o A joint estimation of 4D ash concentration forecasts and, simultaneously, an optimization of the eruption
source term (column height and vertical mass distribution)
o An ensemble containing a minimum 30 ensemble members
• Build on Parallel Data Assimilation Framework (PDAF; http://pdaf.awi.de/trac/wiki), an open-source
software environment for ensemble data assimilation providing fully implemented and optimized data
assimilation algorithms, in particular ensemble-based Kalman filters like LETKF and LSEIK.
6
Data assimilation cycles
• Data assimilation in FALL3D:
o Step 1: Data insertion from satellite retrievals (this talk)
o Step 2: Ensemble Transform Kalman Filter (EnTKF) modelling in the frame of ChEESE
7
Example 1: Puyehue 2011 (ash)
FALL3d-8.0 model setup
phase 4 June 2011 at 21:00 UTC
phase duration continuous emission
run duration 120h
column h 9 km (a.v.l.)
meteorological WRF model at 4km resolution
model 100 vertical levels
FALL3D 0.1º, 60 vertical layers
resolution (660x350x60)
TGSD estimated from h and viscosity µ
• METEOSAT-10 (SEVIRI) retrievals every 1 hour
• Good dataset but with substantial cloud interference at some time intervals
8
Example 1: Puyehue 2011 ash data insertion
Datainsertion
Data
Data insertion
insertion
at557June
at
at June23:00z
June 07:00z
15:00z
(36h
(8h(18h after
afterafter to)
insertion)
to)
9
Example 2: Raikoke 2019 (SO2)
FALL3d-8.0 model setup
phase 21 June 2019 at 18:00 UTC
phase duration 14h emissison
run duration 90h
column h varying, 4 to 15 km
meteorological GFS model at 0.25º resolution
model 37 vertical levels
FALL3D 0.1º, 80 vertical layers
resolution (600x350x80)
TGSD estimated from h and viscosity µ
• HIMAWARI-8 (AHI) SO2 retrievals every 10 min
• Very good signal with (little) cloud and water vapour interference (?)
10Example 2: Raikoke 2019 SO2 data insertion
Datainsertion
Data insertion
atat23
22June
June03:00z
09:00z
22:00z
(18h(15h
(36h afterafter to)
insertion)
11Conclusions
1. New generation of geostationary satellites provides unprecedented space-time data resolution
2. Data assimilation in FALL3d-8.0 is under development:
o Data insertion improves notably “short-term” model forecasts (up to 24/36h)
o Less-dependency on source term uncertainties
o On-going implementation of on-line assimilation cycles build on PDAF library
3. ChEESE PD#12 targets at HPC-based products/services with:
o Unprecedented resolution
o High-frequency assimilation
12Conclusions
THANK YOU
13You can also read
