sea state from sentinel-1 sar for maritime situation awareness
TRANSCRIPT
Sea State from Sentinel-1 SAR
for Maritime Situation Awareness
09.10.2019
Andrey Pleskachevsky
Björn Tings
Sven Jacobsen
1. Introduction
2. New algorithm for S1 WV
3. Algorithm function
4. Ground truth and cross validation
5. Conclusion
German Aerospace Center (DLR)
Maritime Safety and Security Lab, Bremen
1. Introduction
2. New algorithm for S1 WV
3. Algorithm function
4. Ground truth and cross validation
5. Conclusion
1.1. Introduction
German Aerospace Center (DLR)
Maritime Safety and Security Lab, Bremen
(Team SAR Oceanography)
Remote Sensing Technology Institute
SAR Signal Processing
- Development of algorithms
- Prototype software for NRT
- Processor software for ground station NZ
- Processing data – algorithm improvements,
geophysics
German Aerospace Center (DLR)
National Ground Segment, Neustrelitz
German Remote Sensing Data Center
Ground station “Neustrelitz”
NRT services
Integrated Processor for MSA: Near Real Time services(NRT)
1.2. Concept: maritime situation awareness (MSA) for safe navigation
ice classification
oil detection
ship detection
sea state
surface wind
coastline
fast, full automatic
raster processing
layer processing
information from different
layers helps each over
and improve accuracy
coupled processor
DLR Maritime safety and security Lab Bremen
algorithms and processor development DLR Ground Station Neustrelitz (NZ)
NRT chain
NRT products
Map client, ftp, E-mail
FUSION
WITH DATA
FROM OTHER
SOURCES
+ measurements
+ forecast
+ ship AIS
SAR image
operationally:
-sea state
-wind
-ships
-icebergs
Screenshot of the NRT Service demonstrator at DLR Ground Station “Neustrelitz”
1.3. Sea State Processor for Maritime Situation Awareness
NRT services: SENTINEL-1 IW daily processing for German waters - waves, wind, ships
Raster: 6 km, Subscenes: 2.5kmx2.5km
Different product layers available on the GeoServer in NRT and
displayed on the Maritime Security Web-mapping Client.
DLR Ground Station
NRT chain
Arctic Sea, 05.01.2017
1.4. Support of a research cruise in Antarctic Seas
Processed in NRT
and send to research
vessel
“Akademik Treshnikov” on
Antarctic Circumnavigation
Expedition
route optimization International Antarctic
Circumnavigation 2017
1.5. Storm Tracking
Hurricane „Irma“ 2017 (S-1)
Hurricane Irma
NASA:
NOAA's GOES-East visible image
10.09.2017 01:08
KUBA
FLORIDA
2017-09-10 23:25 UTC
new techniques and
algorithms allow observation
and validation of forecast
models worldwide
ESA news:
Sentinel-1 sees through hurricanes
“… information about the sea state can
help to assess how destructive a
hurricane is and can predict its path
respectively time and location on which
it will make landfall ….”
estimation of sea state fields
from Sentinel-1 IW in NRT
75cm RMSE world wide
55cm North Sea, Baltic, etc.
1. Introduction
2. New algorithm for S1 WV
3. Algorithm function
4. Ground truth and cross validation
5. Last slide
Sentinel Wave Mode WV
20km × 20 km vignettes each 200km
wv1 incidence angle ~23°
wv2 incidence angle ~34°
► ~120 orbits/month collocated with NDBC buoy < 50km
~ 130.000 GB/month
~ 1.5 TB /year
Ground truth:
► model data:
temporal interpolated 3h outputs :
► CMEMS (1/12°)
► WW3 NOAA (0.5° >interpolated)
► measurements:
NDBC collocated stationary 41 buoys (Atlantic & Pacific)
2.1. Data general information
Scientific Points:
- non linear SAR imaging of moving waves
- different sea state systems – different SAR imaging
- accuracy of 50cm – 10.000 combinations of waves/dir/wind
- only ~50% of imagettes have no distorted wave-looking structures
► S1 WV overfly:
30 – 160 imagettes
2.5 – 15 GB
S1 data general information
WV SLC products
2.2. Algorithm general information
Total integrated parameters (1):
- SWH (Hs) significant wave height
- Tm1 first moment wave period
- Tm2 second moment wave period
- Tm-1 weighted mean wave period (Te)
Partial integrated parameters (2):
- SWHswell dominant swell significant wave height
- SWHwindsea windsea significant wave height
- Twindsea windsea wave period
Parameters estimated (two groups)
0.40m / 0.45m
0.40m / 0.45m
0.62s / 0.66s
wv1 / wv2 RMSE
0.34m / 0.38m
0.51s / 0.56s
0.46s / 0.51s
0.62s / 0.66s
- Tp peak wave period
- Tswell swell period
Algorithm for S1 WV: CWAVE_EX (Extended)
bias for all parameters ~0
1. Reading
Sea State Processor
2. Calibration
3. Land masking
5. Parameters (features) estimation:
primary and secondary
7. Model functions: TerraSAR-X, Sentinel-1 IW, Sentinel-1 WV
8. Results control
9. Results converting to products (interpolation, xml, nc, etc.)
4. Filtering
6. Filtering by using estimated parameters
2.3. Model validation for significant wave height Hs: ~ 35cm
wv1 wv2
2.4. Wave height validation with NDBC buoys: ~ 43 cm
SWH tuned with buoy RMSE=0.34 m Scattering Model/Buoy = 0.31 m (50km < collocations)
model / measurement
2.5. Model validation wave periods Tm1, Tm2, Tmean
RMSE=0.51 s
RMSE=0.55 s
wv1
wv2
Tm1 ~ 0.52 sec
RMSE=0.51 s
RMSE=0.52 s
wv1
wv2
Tm2 ~ 0.51 sec
RMSE=0.66 s
RMSE=0.62 s
wv1
wv2
Tmean ~ 0.64 sec
1. Introduction
2. New algorithm for S1 WV
3. Algorithm function
4. Ground truth and cross validation
5. Last slide
3.1. Empirical function parameters
Wind 2. Geophysical
Parameter type Parameters first order
NRCS, Norm.-variance, skewness, kurtosis,
+ 5 additional parameters (will be published later)
1. Subscene properties and statistics
3. GLCM (grey level co-occurrence matrix) GLCM-mean, variance, entropy, correlation, homogeneity, contrast,
dissimilarity, energy
5. Spectral-B 20 parameter by using orthonormal functions,
cutoff by ACF (autocorrelation function)
using spectral rings for different wavelengths 4. Spectral-A
Goda-parameter, Longuet-Higgins-parameter,
+ 5 additional parameters (will be published later)
Function: linear regression 𝑃𝑖 = 𝐴𝑛𝑆𝑛
𝑁
𝑛=0
Solution: quadratic minimization using SVD (singular value
decomposition) – optimal solution for a linear system
Parameters type 2
Wind, CMOD algorithm Geophysical
3.2. Empirical function parameters
NRCS, Norm.-variance, skewness, kurtosis Subscene properties and statistics
Parameters type 1
Additional 6 parameters will be published later
𝜎0 = 𝐷𝑁
𝐾𝑠2
2
𝑠𝑘𝑒𝑤 =1
𝑛 𝜎0𝑖 − 𝜎𝑜
3/𝑠3𝑛𝑖
𝑘𝑢𝑟𝑡 =1
𝑛 𝜎0𝑖 − 𝜎𝑜
4/𝑠4𝑛𝑖
with s=std(σo)
𝑐𝑣𝑎𝑟 = 𝑣𝑎𝑟𝐼 − 𝐼
𝐼 variance of normalized image intensity
NRCS
skewness - measure of the asymmetry of the
probability distribution about its mean.
kurtosis - measure of the steepness of the
probability distribution of a random variable.
GLCM - Gray Level Cooccurrence Matrix mean, variance, entropy, correl, homogeneity, contrast,
dissimilarity, energy
1. Gray level matrix 64 values for NRCS
2. Frequency matrix C(i,j):
combination of gray matrix values Y and X directions
3. Parameters
GLCM - a tabulation of how often different
combinations of pixel brightness values occur
in an image in certain directions offset by
certain distances
3.3. Empirical function parameters type 3: GLCM
spectra rings, Goda, Longuet-Higgins
Additional 5 parameters (wuill be published later) Spectral
normalized subscene image spectrum
FFT
► spectra rings (domains)
1 - 30 - 75 - 390 - 600 – 2000 >
E30 E75 E390 E600 E2000 E2500
wavelength, m
max
min
max
min
),(x
x
y
y
k
k
xyy
k
k
xIS dkdkkkISE
►whole integrated energy, no-noise int. energy
►energy integrated according to W. Rosenthal (scaled with 1/k2i)
►Longuet-Higgins and Goda parameters
►2 domains inside and outside cut-off
3.4. Empirical function parameters type 4: Spectral-A
Spectral 20 parameters by using orthonormal functions,
+ cutoff by ACF (Auto-Correlation-Function) Parameter Sij = Spectrum_Normalized * Orthonormal function ij
Image spectrum normalized
Orthonormal function ij = GegenbauerPolinomi*harmonic_functionj
orthonormal functions
3.1.4. Empirical function parameters type 5: Spectral-B
1. Introduction
2. New algorithm for S1 WV
3. Algorithm function
4. Ground truth and cross validation
5. Last slide
Hawaii (groop-2)
51002
51004
51202
51208
Alaska, Canada (groop-1)
46001
46004
46005
46035
46036
46066
46073
46075
46078
46085
46184
46208
46246
4.1. Collocation < 50 km NDBC stationary buoys: 3 groups
Atlantic (groop-3)
41001
41002
41041
41044
41046
41047
41049
44008
44011
44066
44095
44137
44139
44150
Collocation < 50km – < 65 km from imagette center
collocated
imagette
Listing S1 WV relative orbits collocated with NOAA NDBC buoys <50 km. Each orbit A or B ~2/3 times /month
Pacific (Canada, Aleutian islands, Hawaii) Atlantic
4.2. Collocated S1 WV orbits (collocation <50 km)
Ideal case:
~36 collocations vw1 → ~ 80 collocation/month → ~ 960 collocation/year
~37 collocations vw2 → ~ 80 collocation/month → ~ 960collocation/year
Reality:
~600 wv1 /year
~600 wv2 /year
collocation potential
Comparison WW3 and CMEMS (at position NDBC 46001)
Temporary: 3h output → 20min interpolated
Spatially: WW3 on 0.5deg →0.25deg
CMEMS
1/12°
WW3 NOAA
1/2°
2018-01-01 00:00
2018-01-01 00:00
Conditions for alghorithms training:
Latitude: -60° < LAT < 60°
Variance: wv1: 1.0 < nv < 2.0 (variance of normalized SAR image)
wv2: 1.0 < nv < 1.4
model collocations for buoy-collocated S1 WV orbits:
~90 orbits
~ 6000/6000 imagettes ww1/wv2 per month
4.3. Model data: two independent models CMEMS and WW3-NOAA
4.4. Cross validation SWH: two models and NDBC buoys
wv1 / wv2 RMSE (m)
VALIDATION DATA
TUNING DATA CMEMS WW3 BUOYS
CMEMS 0.33 / 0.38 0.35 / 0.40 0.42 / 0.44
WW3 0.34 / 0.39 0.34 / 0.39 0.44 / 0.46 WW3 & CMEMS 0.34 / 0.39 0.43 / 0.45
CMEMS
Tuning
Validation
WW3 Tuning
Validation
WW3 & CMEMS Tuning
Validation
BUOYS
Validation
CMEMS
Validation
WW3
Validation
Validation
Validation
1. Introduction
2. New algorithm for S1 WV
3. Algorithm function
4. Ground truth and cross validation
5. Last slide
Processing: 1 task 20 processors ~ 5-15 min for one orbit of 3 GB – 15 GB (first 2-6min reading)
Algorithm: CWAVE_EX
Data: S1 Wave Mode (WV)
Developed: DLR
Accuracy