• State of Art in EO of Inland Waters: a UK perspective Andrew Tyler (University of Stirling)

1. State of the Art in the EO of Inland Waters:
A UK perspective
Andrew Tyler
IUKWC Wokshop June l 2017
Enhancing Freshwater Monitoring Through Earth Observation

2. Inspired by opportunity and need…
Lake Balaton, Hungary Landsat 7
Need to monitor for management, protection and resilience
Recognition of the spatial and temporal heterogeneity
Tendency for reactive monitoring
Scale of the problem
Algorithm stability
Challenges of optically complex waters
Growing capacity and capability of satellite platforms
Tyler et al., 2006, IJRS

3. Airborne Hyperspectral: PC retrieval
IUKWC June 2017: Enhancing Freshwater Monitoring Through Earth Observation
Hunter, et al. (2008). RSE
Hunter, et al. (2008). Limnol. Ocean
Hunter, et al. (2009). Envi. Sci. Tech.
NERC ARSF:
AISA Eagle and Hawk

4. MERIS in inland waters
IUKWC June 2017: Enhancing Freshwater Monitoring Through Earth Observation
Envisat: MERIS

5. Inner basin
48.4 mg Chla m-3
Middle basin
13.1 mg Chla m-3
Outer basin
1.00 Chla mg m-3
Elterwater
Lat: 54.4273 / Long: -3.0230
Loughrigg Tarn
STIR: 27.8 mg m-3
EA: 27.2 mg m-3
World View-2: Water optical type classification
Turbid NIR-red ratio
R754/R659
Clear green-blue ratio
R546/R478
In-water algorithm
Mapped
Level-2
Chla
Hunter and Tyler (2012) EA Report

6. Inner basin
48.4 mg Chla m-3
Middle basin
13.1 mg Chla m-3
Outer basin
1.00 Chla mg m-3
Loughrigg Tarn
STIR: 27.8 mg m-3
EA: 27.2 mg m-3
IS R(608) > 0.04
YES:
Chla ~ R(754)/R(659)
NO:
Chla ~ R(546)/R(478)
Hunter and Tyler et al. (2012) EA Report Report
With Citizen Science Based Validation

7. EO Challenges – Global Scale
IUKWC June 2017: Enhancing Freshwater Monitoring Through Earth Observation
EO key challenges:
 Diversity in retrieval algorithms & validation approaches
 Inland water remote sensing community appears fragmented
Search Keys: remote sensing, water quality,
lakes (2015)
Filter: use of in-situ data for
development/validation
Number of lakes per publication:

8. Our approach
Lake ecology & modelling
Global lakes observatory
for 1000 study lakes
Environmental statistics
EO lake water
quality
EO + modelled
catchment
drivers
EO lake water
temperature
Change over time
(within / between lakes)
Attributing drivers
of environmental
change
Time-series data &
web visualization
IUKWC June 2017: Enhancing Freshwater Monitoring Through Earth Observation

9. A Global Partnership: LIMNADES
www.limnades.org
o data from almost 1500 inland systems
o radiometric data ~4000 stations >250 lakes
o at least 40 peer-reviewed papers
IUKWC June 2017: Enhancing Freshwater Monitoring Through Earth Observation

10. A Global Partnership: LIMNADES
Parameter Units Range Median Lakes
Chla mg m-3 0.03-13296.70 12.34 208
PC mg m-3 0 – 24677 28.79 60
TSM mg L-1 0.09-2533.30 10.54 81
ISM mg L-1 0.01-359.42 13.35 10
aCDOM(442) m-1 0.004-42.467 0.8206 83
aph(442) m-1 0.036-454.976 0.5148 19
aph(442)/[Chla] m2 mg-1 0.002-0.257 0.0156 19
aNAP(442) m-1 0.004-12.540 1.3875 23
aNAP(442)/[TSM] m2 g-1 0.001-0.410 0.0573 19
bp(532) m-1 0.524-17.788 2.2442 6
bp(532)/)/[TSM] m2 g-1 0.136-1.838 0.5872 6
bbp(532) m-1 0.004-0.292 0.0676 9
bbp(532)/)/[TSM] m2 g-1 0.0003-0.1003 0.0118 9
IUKWC June 2017: Enhancing Freshwater Monitoring Through Earth Observation

11. OWT classification
• ~4,000 Rrs spectra
• K-means clustering
• Optimum number of
clusters determined
statistically
Spyrakos et al., submitted ….
IUKWC June 2017: Enhancing Freshwater Monitoring Through Earth Observation

12. OWT classification
OWT Dominant characteristics
OWT1 Hyper-eutrophic waters with cyanobacteria scum and
vegetation-like reflectance
OWT2 Common case waters with diverse reflectance shape and
marginal dominance of pigments and CDOM over inorganic
suspended particles
OWT3 Clear waters
OWT4 Turbid waters with high organic content
OWT5 Sediment-laden waters
OWT6 Balanced effects of optically active constituents at shorter
wavelength
OWT7 Highly productive waters with reflectance peak with elevated
reflectance at red/near-infrared spectral region
OWT8 Productive waters with reflectance peak close to 700 nm
OWT9 Optically neighbouring to OWT2 waters but with higher
reflectance at shorter wavelengths
OWT10 CDOM-rich waters
OWT11 High in CDOM waters with cyanobacteria presence and high
absorption efficiency by NAP
OWT12 Turbid, moderately productive waters with cyanobacteria
presence
OWT13 Very clear blue waters
IUKWC June 2017: Enhancing Freshwater Monitoring Through Earth Observation
Spyrakos et al., submitted ….

13. OWT classification
Lake Balaton, HungaryTONLÉ SAP Lake, Cambodia
IUKWC June 2017: Enhancing Freshwater Monitoring Through Earth Observation
Spyrakos et al., submitted ….

14. Algorithm validation
Step 1: Validation
of original
algorithms
Step 2: OWT
cluster-wise
algorithm tuning
Step 3: Algorithm
selection per
cluster or cluster
family
LIMNADES in situ
MERIS matchups
Optical Water Types
IUKWC June 2017: Enhancing Freshwater Monitoring Through Earth Observation

15. Algorithm validation
IUKWC June 2017: Enhancing Freshwater Monitoring Through Earth Observation

16. In situ algorithm validation
Y=0.7692x + 0.4175
R2=0.698
MAPE=164.5
Y=0.7863x + 0.311
R2=0.7891
MAPE=84.67
Y=0.8313x + 0.2197
R2=0.7969
MAPE=92.96
• Retuning per cluster
reduced uncertainty
compare to original model
• Further improvements
evident when using the
best performing algorithm
per cluster.
Gurlin 2011 original output Gurlin retuned per cluster
Dynamic algorithm selection per cluster
IUKWC June 2017: Enhancing Freshwater Monitoring Through Earth Observation
Neil et al., in prep….

17. MERIS Calimnos match-ups
Atmospheric correction Polymer, Scape-M, CoastColour, Fub, Boreal Lakes, Megs
Constituent Retrieval FLH, C2R, BL, EUL etc.
Constituent Retrieval (Chla, TSM, CDOM, PC)
cloud OR cloud shadow OR snow_ice OR Glint (Idepix)
Input from initial in-situ
validation to screen out
poorly performing algo.
Standardisation of Rrs and OWT-membership function
3x3 sigma-filter
Ground data
Chla (>50000 spec or
HPLC from ~2000
inland water systems),
TSM (8760 from ~500
water systems), CDOM
(2000 from 78 inland
water systems), PC (532
from 48), in-situ Rrs
(3000 from 250)
Scores
Quality flags
MERIS Algorithm validation
IUKWC June 2017: Enhancing Freshwater Monitoring Through Earth Observation

18. Algorithm performance
Slope
Normalised score [1]
Correlation r
Normalised score [2]
IUKWC June 2017: Enhancing Freshwater Monitoring Through Earth Observation
Spyrakos et al., in prep ….

19. Water type
family
Optical water types Suggested algorithm
1 3; 9; 10; 13 OC2-like
2 2; 8; 11; 12 Rrs708/Rrs665
3 1; 4; 5; 6 Gons, 2005
4 7 QAA (Mishra et al., 2013)
Dynamic Alogorithm Selection
⎯ a dynamic
processing chain for remote sensing
of inland water quality
IUKWC June 2017: Enhancing Freshwater Monitoring Through Earth Observation

20. Operationalisation
IUKWC June 2017: Enhancing Freshwater Monitoring Through Earth Observation

21. Some reflections
The Razelm lagoon system, Romania
• Monitoring ~ 50 % of world’s freshwater – ESA
Sentinel 3 (OLCI) and Sentinel 2 (MSI)
• Dynamic algorithm selection based on OWTs
provides more accurate products
• Processing chain operational for a continuum of
optically complex River-Sea Systems (DANUBIUS-RI)
• Data now being used to interpret drivers controlling
lake status and change
• Continuing need for high quality in situ matchup
data for EO, currently are limited
• biogeochemical >> radiometry
• Chla > TSM >> CDOM > PC > Kd
• GloboLakes simply not possible without community-
wide collaboration
IUKWC June 2017: Enhancing Freshwater Monitoring Through Earth Observation

22. Thank you!
Andrew Tyler
Professor of Environmental Monitoring
Biological and Environmental Sciences
University of Stirling
t +44 1786 467838
e a.n.tyler@stir.ac.uk
w www.stir.ac.uk
w www.globolakes.ac.uk
w www.limnades.org
follow @globolakes
Acknowledgements
Natural Environment Research Council U
ESA Diversity II
Over 30 data contributors around the w