Harmonic & EOF analysis for North Sea SPM...
Transcript of Harmonic & EOF analysis for North Sea SPM...
Harmonic & EOF analysis for North Sea SPM patternsGerben J. de Boer (Deltares, TU Delft)Meinte Blaas (Deltares)Loana Arentz (Deltares)Niels Kinneging (Rijkswaterstaat WaterDienst)Ackowlegdements: Liege GHER group for DINEOF
Contents:1. The context2. Southern North Sea physics3. Harmonic analysis4. DINEOF analysis5. Hybrid analysis
land reclamationsand mining
wind farms
image credit: NASA MODIShttp://www.spaceref.com/news/viewsr.html?pid=23750
1. The context: stability of combined situ + RS signal
• Dutch government (Rijkswaterstaat) is monitoring North Sea SPM and Chl-a > 25 yr• Network of 80 (in 1980s) to 20 (currently) fixed ship-served locations• Visited bi-weekly to bi-monthly with buckets (chemicals)
• Current needs• Impact monitoring for new developments requires more resolution
> wind farms, land reclamation, sand mining• National program for OSPAR, WFD, MSFD requires optimization• Combining and changing observation techniques: ship buckets > buoys + scanfish + EO• RS from different L2 providers and satellites
• Use Remote Sensing data for monitoring too, ensuring consistency and veracity?
# pixe
ls in 1
year
of MERIS data
(binned to cu
rvi-linear
model)
50 to
80 good im
ages
per yea
r
# pixe
ls in 25
years
of in si
tu data
max. 2
6 good sa
mples per
year
1. The approach
Our problem• No in situ match-ups due to low bi-weekly sampling freq.• Gap filling (Nechad et al, 2011) for artificial match-ups won’t help here
Our aimProvide a reduced-order description of the state of the system from EO & in situ• Filtered historic sets to quantify consistency (e.g. cross comparison)• Statistical model to quantify veracity (plausibility) of new samples
Study case• 1 (to do 6) yr Envisat MERIS RR• VU-IVM / Water Insight HYDROPT retrieval
(SPM, Chl-a, CDOM, …)vd Woerd & Pasterkamp, ‘08; Peters et al., ‘05, ‘08
• Binned to curvilinear >1 km grid, focus coastal area
TechniquesHarmonic & DINEOF analysis of SPM• SPM variance quantification• SPM bulk statistic for consistency• SPM patterns for physical insight
Mean SPM
Pietrzak, de Boer, Eleveld, 2011, CSR 31(6). Mechanisms controlling intra-annual mesoscale variability of SST and SPM in the southern North Sea
fine bedsediments
Fortnight + tidalhaline stratifiedRhine ROFI /Coastal riverplume barocliniccoastaltrapping ofwater + SPM
Front: EastAnglia plume+ Frisian front
Seasonally,thermallystratifiedcentral N. Sea
Shallow mixedDogger bank
Central North Sea watersEast Anglia water
Central Channel waterContinental waters
2. Physics Southern North Sea
3. Pure harmonic analysis
Annual cycle:coherent phases: significant
Spatial modePhase annual cycle
Jan.
Oct.
Apr.
1
summer Thides bottom
SPM
winterstormsstirbed
Temporal mode
winterstormsmixS
Spatial modeAmplitude annual cycle
3. Pure harmonic analysis
Semi-annual cycle (6 month cycle):partly coherent phases:
significant?
1
bloom?
Temporal mode
Spatial modePhase semi-annual cycle
Spatial modeAmplitude semi-annual cycle
3. Pure harmonic analysis
Quarter-annual cycle (3 month cycle):incoherent phases: insignificant
1
Temporal mode
Spatial modePhase quarter-annual cycle
Spatial modeAmplitude quarter-annual cycle
4. Pure DINEOF analysis
EOF mode 1• rather steady: equivalent of A0 from harmonic analysis• max. 9 modes are found
1
Temporal modeSpatial mode
4. Pure DINEOF analysis
EOF mode 2• seasonal cycle, but not sinusoidal
• strong jump downward in spring• bumpy trend upwards in autumn
• spatially quite uniform value
1
Temporal mode
Spatial mode
summer Thides bottom
SPM
winterstormsstirbed
winterstormsmixS
4. Pure DINEOF analysis
EOF mode 3• high frequency behavior ~ 2 months• no obvious relation with physical cycles• spatially heterogeneous: continental
water mass stands out with lower response
1
?
Temporal modeSpatial mode
3+4. Explained variances
• East Anglia plume fits well bear UK• Rhine ROFI fits best with harmonic (3 modes)• DINEOF overall has a better fit (all 8 modes)
ExplainedDINEOF variance
Explainedharmonic variance
summer Thides bottom
SPM
winterstormsstirbed
winterstormsmixS
• DINEOF run 5 times• each time with 1 more harmonic component peeled off
• The harmonic annual mean overtakes EOF1• The seasonal cycle does not take over any EOF mode (EOF2)• Seasonal cycle only gets some significance when higher
harmonics are added: to account for steep spring changes?
5. Peeling off harmonics: explained variances
i i
var
mode mode raw raw
rawpixel wise, discarding clouded pixels
5. Temporal EOFS of harmonic residual
(3rd): 3.9 % seasonal sine
(4th): 2.5 % semi-seasonal sine
(2nd): . . . . . . . . . . . . . . . . 6.9 % EOF2
(1st): 44.3 % harmonic mean (EOF1)
(4th): . . . . . . . . . . . . . . . . . . 2.4 % EOF3
5. EOF2 unexplained pattern
no14.8%
- … - harmonic yr9.2%
- … - semi-year7.9%
- … - quarter-year6.9%
- mean13.6%
5. EOF3: unexplained pattern
no3.2%
- mean2.7%
- … - semi-year2.4%
- … - quarter-year2.4%
- … - harmonic yr2.2%
Conclusions: use harmonic + DINEOF
Modes in term of capture variance:• EOF1 ~ temporal (harmonic) mean
• Contains most of variance: close to 50%• Remove harmonic temporal mean before applying DINEOF• Harmonic A0 is least-squares fit: better than geometric mean / median
• EOF2 has 2nd most variance• Captures a little seasonality from raw EO data• Has only 2-month modes left after removal harmonic season from EO• Tele connection: Rhine ROFI & German Bight• Unexplained physical spatial pattern left with ~ 2-month cycle: more research
• Harmonic seasonal has 3rd most variance• mode only captures variance when higher harmonics are added too:• due to steep spring changes seasonality is not sinusoidal
• higher modes not recommended for reduced state
• Harmonic + DINEOF analysis together can yield reduced state with 3 modes• Plan: Match reduced modes with in situ DINEOF (only 1 mode found, data scarcity)
Questions?
image credit: NASA MODIShttp://www.spaceref.com/news/viewsr.html?pid=23750