New diagnostics to assess the impact of satellite constellation for (sub)mesoscale applications Complementarity between SWOT and a large constellation of pulse-limited altimeters

1
New diagnostics to assess the impact of
satellite constellation for (sub)mesoscale
applicationsComplementarity between SWOT and a
large constellation of pulse-limited altimeters

• M.I.Pujol, G.Dibarboure, G. Larnicol (CLS)
• P.Y.Le Traon, P.Klein (IFREMER),

2
Introduction

• SWOT will provide an unprecedented sampling
  capability by 2019
• Iridium-NEXT telecommunication constellation
  renewed, starting from 2015
• Iridium satellites can take payloads of
  opportunity
• It is technically possible to have AltiKa-like
  pulse-limited altimeters on Iridium-NEXT
• The constellation itself would have intrinsic
  advantages (very cost efficient, temporal
  sampling, robustness vs failures, near real
  time)
• But a constellation of traditional sensors cannot
  replace SWOT images
• What could be the benefits of having a
  constellation of 6 Iridium-NEXT altimeters
  (upcoming missions) in addition to SWOT for
  (sub)mesoscale retrieval ?
• Are Lagrangian diagnostics relevant for this
  study ?

3
OSSE approach

• Protocol
• Reality Earth Simulator outputs from Ifremer
  mesoscale and submesoscale
• Simulate observation by remote sensor (with
  error)
• Reconstruct  observed ocean topography  from
  profile/swath observations using optimal
  interpolation mapping (DUACS center to generate
  the AVSIO products)
• The difference between the reality and the
  observed state is the sum of
• Remote sensing sampling weaknesses (blind spots)
• Remote sensing measurement errors
• Reconstruction imperfections (e.g. oversmoothing)
• Error is always measured in percentage of the
  reality signal variance

4
Simulation details

• Configurations studied
• Classical nadir constellation 3 x nadir
  altimeters (Jason-CS, Sentinel3-B, S3-C)
• SWOT alone
• SWOT 3 altimeters
• SWOT 11 altimeters (6 Iridium, Jason-CS, S3-B,
  S3-C, HY-C, GFO2)
• Error levels optimistic (both on SWOT and nadir
  altimetry)
• only noise and residual roll for SWOT (after good
  cross-calibration)
• 1 cm noise for nadir (radiometer, dual
  frequency/Ka, and good POD)
• Reconstructing the topography at each time step
  and position
• Straightforward optimal interpolation (no model
  assimilation) derived from DUACS tools
• Mapping 1 standard DUACS mapping 100 km / 10
  days (mesoscale)
• Mapping 2 2-step optimal down to 30 km and 5
  days (small mesoscale / submesoscale)
• Regional reconstruction (not just within the
  swath ? temporal coherency analyzed)

5
Ocean reality

• One year of Earth Simulator from Ifremer (Klein
  et al) ? mesoscale and submesoscale
• Theoretical model can be  projected  to any
  region or bathymetry configuration
• ? North Pacific at two locations 38N,210E
  and 45N,210
• RMS of the sea surface height anomaly

lt 100km
TOTAL
gt 100km
6
Instantaneous observations (typical snapshot)
11 x Nadir (Iridium 6 Jason-CS GFO2 HYC S3A
S3B)
3 x Nadir
1 x SWOT
1 x SWOT 11 x Nadir
2 x SWOT
7
Method

• Objective Analysis (OA) method (Le Traon et al,
  1998 Ducet et al, 2000) used for SLA
  reconstruction
• Large and medium mesoscale signal direct OA
  with correlation scales 100km/10days
• Short mesoscale signal 2-step OA method with
  correlation scales 100km/10days and 30km/5days

Map reconstructed with 1/8x1/8 and 3 days
resolution.
? Errors on reconstructed fields are analyzed for
SLA, surface geostrophic velocities (U,V),
vorticity and vertical velocities (W)
8
Illustration of common diagnostics used for OSSEs

• SSH reconstruction error (diff reference
  reconstructed)
• Analysis made at 38N (SWOT temporal sampling is
  optimal)
• Mesoscale SSHA reasonably resolved with 3
  satellites(current applications ofDUACS)
• SWOT alone performslike 4 altimeters
• Adding more sensorsreduces the error but the
  gain is small

SSHA Reconstruction error ( of reality signal
variance)
9
Mesoscale sampling (influence of latitude)

• Only one SWOT sensor ? results change with
  latitude
• Blue is for 38 (optimal temporal sampling 1
  sample every 11 days)
• Grey is for 45 (poor sampling 2 samples in 4
  days, then 18 days with no data)
• Sampling discrepancies disappear when a large
  constellation is added

Delta Time between ascending and descending arcs
on SWOT x SWOT crossovers
45
38
-10 days
10 days
10
Mesoscale sampling (geostrophic velocities)

• Reconstruction error at 45N on U (blue) and V
  (red) components
• Observing true gradients is much more difficult,
  even on  simple  mesoscale
• Second SWOT or constellation ? error divided by a
  factor of 2
• Direct benefit for traditional altimetry
  applications at regional scale

Geostrophic velocities reconstruction error (
of reality signal variance)
11
The lyapunov exponents

• Potential of using Lagrangian metrics to
  charactrise the impact of satellite constellation
  
• Test has been performed using a Lagrangian
  approach with the calculation of the Lyapunov
  exponents (FSLE for finite size Lyapunov
  exponents) of the velocity data set
• ? direct measure of the local stiring
• ? characterise the trajectories of initially
  close particules that are
• quickly separated along the streaching directions
  
• In practice a set of tracers (initially
  separated with a specific distance) are followed
  in time during the advection by the velocity
  field.
• FSLE is the time it takes to the tracers to reach
  a given separation distance
• Ref papers DOvidio et al. (GRL, 2004), DOvidio
  et al. (DSR, 2009)
• DOvidio et al.., (2004) software is rewriting
  and will be available soon.

12
Reconstructing lyapunov exponents
Reality (Earth Simulator)
1 x SWOT 11 x Nadir
3 x Nadir
1 x SWOT
13
Do we need optimal 2-step mapping ?
Reality (Earth Simulator)
SWOT 11Nadir (2-step)
SWOT 11 Nadir (standard)
14
Integrated advection error

• Initial state hundreds of particules to be
  advected ?
• Position analyzed every 3 hours over 5 days
• The mean distance between reference and observed
  trajectories
• gives an estimate of the integrated error
• SWOT alone still has an average error (42km)
  superior to the mapping scales (30km) ?
  observation lacking
• Adding a second SWOT or (better)a constellation
  of 11 nadir reduces theerror by 50 (25km)

Average error on tracer position
5 days
15
Do we need optimal 2-step mapping ?
SWOT (standard) SWOT (2-step) SWOT11 Nadir
(standard) SWOT11 Nadir (2-step)

• If SWOT is alone, the 2-step mapper significantly
  reduces the error at regional scale (optimal
  interpolation uses statistical decay between
  sparse images)
• When a constellation is merged with SWOT, the
  dense 1D profiles can preserve the SWOT 2D
  information until a new swath refreshes the scene
  ? improvements from 2-step mapping is marginal
  (standard maps are good enough if observation is
  not filtered)

16
Conclusions OSSE results

• A large altimetry constellation can complement
  SWOT images
• To fill SWOT temporal gaps between 2D images
  (with dense 1D profiles)
• To fill SWOT observation weaknesses at certain
  latitudes (22-day orbit)
• ? To better observe smaller scales (error divided
  by a factor of 2 for signals gt 30km)
• Optimal 2-step mapping (vs. traditional DUACS
  mapping)
• 2-step is not necessary for SWOTconstellation
  (dense measurements)
• 2-step is needed for SWOT alone to balance the
  sparser temporal observation (standard mapping
  would over-smooth between 2D images)

17
Conclusions on Lagrangian metrics

• Lyapunov exponents useful but qualitative
• further work need to understand the
  fiability/sensitivity of this Lagrangian method ?
  
• (impact of the parameterisation, sensitivity to
  the sampling)
• Quantitative diagnostics with the Integrated
  advection error are satisfying.
• Are Lagrangian diagnostics relevant for a NRT
  monitoring (OSE)?

DUACS régional AMESD
DUACS Global product
18
Conclusions on Lagragian metrics

• Lyapunov exponents useful but qualitative
• further work need to understand the
  fiability/sensitivity of this Lagrangian method ?
  
• (impact of the parameterisation, sensitivity to
  the sampling)
• Quantitative diagnostics with the Integrated
  advection error are satisfying.
• Are Lagrangian diagnostics relevant for a NRT
  monitoring (OSE)
• Where is the truth ? How could we verify that
  the small scale introduced in the field are
  realistic ?
• Is there specific signatures on FSLE/FTLE
  results of some specific signals? (internal
  wave, sampling discontinuity, ..),
• Consistency with tracers like ocean colour ?
• Interest to use Lyapunov exponent for model
  simulations intercomparison and validation ?