Sentinel-2 MSI in the Monitoring of Lakes and Coastal

Sentinel-2 MSI in the Monitoring of Lakes
and Coastal Waters in Finland: Spectral
and Spatial Resolution Considerations
Kari Kallio, Sampsa Koponen, Jenni Attila, Mikko
Kervinen, Timo Pyhälahti
Finnish Environment Institute
Carsten Brockmann, Tonio Fincke
Brockmann Consult
Lena Kritten
Free University of Berlin
Sentinel-2 for Science Workshop, ESRIN, May 20-22 2014
- Global Lakes Sentinel Services
● Collaborative Project (2013-2016) funded by the EU 7th
Framework Programme
● Develops processing tools for the upcoming Sentinel-2 and
Sentinel-3 satellites to monitor lakes and reservoirs.
○ Algorithm development for various water types
http://www.glass-project.eu/
Outline
● Characteristics of lake and coastal waters in Finland
● Spatial resolution analyses
● Potential of MSI band wavelengths for water quality estimation
with band ratios
● Spectral inversion algorithm SIOCS (The Sensor-Independent
Ocean Colour Processor)
Coastal waters and lakes in Finland –
need for high resolution data
Length of coastal line (main land) : 1 100 km.
Number of islands along the Finnish coast: 28 000 (> 0.3 hectar).
 large spatial variation of water quality and a huge monitoring task.
LANDSAT5 TM Archipelago sea, 2 June 2007 Turbidity
10 km
Number of coastal
WFD water bodies is 215.
~ 40% of them cannot be
monitored with 300 m data.
Lakes in Finland
Number of lakes is 56 000 (> 1 hectar)
MERIS pixels
Lakes not
possible to
monitor with
300 m
resolution
instruments
5
WFD monitoring of Finnish lakes with 300 m
data (MERIS, OLCI) – Effect of spatial
resolution
Number of lakes
Total number of WFD lakes (> 0.5 km2) is 4596.
1000
800
300 m sensor with
land mask
695
15%
Land
600
400
200
201
4%
300 m sensor
mask
with
buffered land
1 pixel
mask:
bufferthe mask is
extended by one
extra pixel from
the shore
0
10 pixels
With Sentinel 2: All WFD lakes and many smaller lakes
Water quality algorithms: analysis of
spectral configuration and test of
SIOCS
Based on HydroLight (Version 5.2) simulations
Water quality input to HydroLight:
• Set 1: Stepwise data based on water quality distributions of
Finnish lakes (N=3375) for SIOCS training
• Set 2: Extensive algorithm testing (band ratios and SIOCS):
based on measurements in 5553 stations
SIOPs for HydroLight from Finnish lakes
Algorithms
• ‘Classical’ band ratios (Landsat8 OLI, MSI, OLCI)
• Spectral inversion SIOCS (MSI)
Spectral inversion algorithm: Sensor
Independent Ocean Color prosessor (SIOCS)
● Joint project of Brockmann Consult and Free University of
Berlin.
● Sensor bands can be selected
● User can use own IOPs/SIOPs and concentration ranges
● Planned to be available in the BEAM EO toolbox
● SIOCS is under development. SYKE has participated in
testing of SIOCS and used HydroLight in its training
8
Spectral resolutions of OLI, MSI
and OLCI in 390-715 nm
4
Landsat8 OLI 3
S2 MSI2
S3 OLCI1
0
400
450
500
600
550
wavelength nm
650
700
MSI 650-680 nm: 2. absorption maximum of phytoplankton
MSI 698-713 nm: small absorption by particles and CDOM
750
SIOCS: main components
Test with Finnish lakes using the S2-MSI bands
IOPs (aph,
aCDOM, ad,
bTSM) for the
bands
3375 cases,
stepwise
variation
Rrs measured
Rrs simulations
HydroLight
simulated,
5553 cases,
based on
measured
concentrations
HydroLight
LUTs
Inversion operator
- cost function
- stop criterium
IOPs at 443 nm
SIOPs
Concentrations
Finnish lakes
10
Chl-a: band ratios and SIOCS
SIOCS S2-MSI
SIOCS
MSI Chl-a
S2-MSI
MSI Chl-a
CHL measured µg/l
40
20
0
0.5
1
Rrs (698-713)/Rrs (650-680)
R2 = 0.99
N = 5059
80
60
40
20
0
1.5
0
20
40
60
CHL SIOCS µg/l
80
S3-OLCI
OLCI Chl-a
4.37
80 y 2= 35.2x
R = 0.81
N = 5498
60
Chl-a
Chl-a
4.81
80 y 2= 27.0x
R = 0.81
N = 5498
60
40
20
0
0.5
1
Rrs (704-714)/Rrs (660-670)
1.5
11
TSM: single band and SIOCS
SIOCS
MSI
SIOCS
S2-MSI
TSM
S2-MSI
MSI TSM
TSM
TSM measured mg/l
R2 = 0.98
N = 5497
30
20
10
0
0
0.005
0.01
0.015
20
10
0
0.02
R2 = 1.00
N = 5059
30
0
-1
Rrs (698-713) sr
Landsat8-OLI
TSM
OLI
TSM
TSM
R2 = 0.99
N = 5497
30
20
20
10
10
0
30
S3-OLCI
TSM
OLCI
R2 = 0.89
N = 5497
30
10
20
TSM SIOCS mg/l
0
0.005
0.01
0.015
(640-670)-1sr-1
R(640-670)
Rrs
sr
rs
0.02
0
0
0.005
0.01
0.015
0.02
-1
Rrs (704-714) sr
12
CDOM: band ratios and SIOCS
SIOCS
S2-MSIMSI
aCDOM(443)
SIOCS
(443) measured 1/m
y = 5.16x 1.29
R2 = 0.97
N = 5498
15
10
0
10
CDOM
5
0
1
2
3
Rrs (705)/R
(560)
Rrs(698-713)/R
rs(542-578)
rs
R2 = 0.98
N = 5059
15
a
a
CDOM
(443) 1/m
MSI
S2-MSI
CDOM
5
0
4
0
5
10
15
aCDOM(443) SIOCS 1/m
S3-OLCI
CDOM
OLCI
15
(443) 1/m
y = 5.77x 1.23
R2 = 0.98
N = 5498
CDOM
10
5
0
a
a
CDOM
(443) 1/m
Landsat8-OLI
OLI CDOM
0
1
2
R
(665)/R
(560)
Rrs(640-670)/R
rs
rsrs(530-590)
3
y = 3.34x 1.86
R2 = 0.96
N = 5498
15
10
5
0
0
1
2
3
R
(709)/R
(560)
Rrs(704-714)/R
rs
rsrs(555-565)
4
13
Validation measurements in Finland
Moored automatic stations (Chl-a, turbidity)
• Coastal: 2-3 stations, lakes 3 stations
Ship-of-Opportunity, Baltic Sea(Alg@line)
• Chl-a, turbidity
• Rrs
Lakes
• Rrs, IOPs
SIOP/IOP datasets
• Lakes: Ylöstalo et al. 2014 RSE 148: 190–205
• Baltic Sea: data analyses in progress
• River influenced coastal waters: to be measured
S2-MSI data will help in
•
WFD ecological classification (Chl-a, transparency)
•
Estimation of the impact of water protection
measures (Chl-a, turbidity)
•
Mapping of areas influenced by river plumes
(turbidity)
•
Macrophyte mapping (e.g. common reed belts)
•
Bottom refrectance may become a limiting factor in
part of the coastal and lake environments
Conclusions
•
High resolution data of S2 MSI can cover all the Finnish
coastal waters and lakes.
•
S2 MSI improves the estimation accuracy of TSM
compared to the currently operational HighRes sensors
and is likely to enable the estimation of Chl-a.
•
SIOCS will be improved e.g. by testing different options for
the initial guess values.
•
The SIOCS spectral inversion algorithm will be very useful
 enables e.g. local SIOP adjustment  improved
regional/local products.
Thank you!