3.1 IUKWC Workshop Freshwater EO - Stefan Simis - Jun17
1. Progress and challenges towards a
global near real-time inland water
quality observation service
Stefan Simis
Senior Earth Observation Scientist (Inland waters)
stsi@pml.ac.uk
2. Challenges for a global inland water quality
observation service
• Many geographically sparse locations
• Water types vary between and within water bodies
• Different water types require different water constituent retrieval algorithms
• Third party algorithms: no consistent programming language
• Validation data are sparse, not readily available for new sensors
• Output formats and interfaces suitable for small and large scale users
The Calimnos (‘good lake’) processor:
- Designed to address (most of) these challenges efficiently.
- Contains algorithms tuned to various water types
- Can be extended with other water types, algorithms, sensors
- Currently working with ENVISAT MERIS (2002-2012) data
- Being readied for Sentinel-3 OLCI (2016 ->) data
3. Challenge: many geographically sparse locations
NERC GloboLakes
time-series 2002-2012
1000 lakes
‘linking water quality to
environmental change’
Algorithm development
and validation
Copernicus Global
Land Service (CGLOPS)
Lake Water
10-day operational public
service: reflectance,
temperature, turbidity,
trophic state index
Available 2018
4. Challenge: many geographically sparse locations
Sensor n files /
scenes /
passes
Daily Volume
MERIS (300m)* ~60 ~60 GB
S3-A OLCI (300m) ~220 ~130 GB
S2-A MSI (10-60m) 4204 ~2.1 TB
Daily volume of satellite data Processing whole satellite passes (pole to pole)
requires large processing (100s PCs) and
storage capacity.
Solution for 300m resolution data:
- Download (nearly) all passes
- Subset areas around individual lake
- Process in parallel and store output
- Aggregate results at desired interval
Daily computing <10 PCs for 1000 lakes:
• scalable and transferable
• Fast archive reprocessing (algorithm testing)
However:
• all algorithms must accept spatial subsets
• Sentinel-2 processing will require global data
access, optionally cloud-based approach
• S2 and S3 volumes to double when
B-satellites become operational
*136 TB in 10-yr archive
5. Challenge: diversity in water types, algorithms
Type Model Reference
(Semi-)empirical
NIR-red BR
MERIS 2-Band
708/665
Gilerson et al. 2010
Gurlin et al. 2011
Gons et al. 2005
MERIS 2-Band
753/665
Gilerson et al. 2010
Gitelson et al. 2011
Moses et al. 2009.
MERIS 3-Band Gitelson et al. 2008
Gitelson et al. 2011
Gurlin et al. 2011
Moses et al. 2009
MERIS NDCI Mishra et al. 2012
Empirical OC MERIS OC2E
MERIS OC3E
MERIS OC4E
O’Reilly et al. 2000
Neural Network NN_Chl
NN_IOP
Ioannou et al. 2013
Analytical MERIS QAA [Turbid] Mishra et al. 2013
MERIS GSM Maritorena et al.
2002
MERIS Matrix
Inversion
Boss & Roesler 2006
Peak Height
Method
MPH Matthews et al. 2012
Type Model Reference
Empirical Binding red
Zhang 708
Vantrepotte
665
POWERS 560
Binding et al. 2006
Zhang et al. 2010
Vantrepotte et al. 2011
Eleveld et al. 2008
D’Sa 665/560
Dekker
490,560
Dekker
560,665
D’Sa et al. 2007
Dekker et al. 2002
Loisel 3-Band Loisel et al. 2014
(Semi-)
Analytical
Binding A
Nechad 665
Nechad 681
Nechad 708
Nechad 753
Binding et al. 2010
Nechad et al. 2010
Type Model Reference
Empirical Duan 709/620 Duan et al. 2012
Duan 3-Band
Song 3-Band
Duan et al. 2012
Song et al. 2013
(Semi-)
analytical
Mishra QAA 13
Mishra QAA 14
Simis NBR
Mishra et al. 2013
Mishra et al. 2014
Simis et al. 2005
Chlorophyll-a
PhycocyaninTSM
6. 13 optical water types…
Lake Balaton
Tonlé Sap Lake
…mapped to time series of MERIS imagery (2002-2012)
Lake Balaton, Hungary
Lough Neagh, Northern Ireland
In situ
Satellite
…each mapped to specific algorithms
…to produce time-series without discontinuities
7. Highest density of observations for inland waters ever.. But still needs careful screening
9. POLYMER
GloboLakes OWT
optical
water types
[LIMNADES
2016]
Reflectance algorithms
algorithm
mapping
Water
Reflectance
wavebands
passes
database
S-3
[SAFE]
MERIS
[FSG/SAFE]
ESA
catalogue
Discover
Download
Ingest
Subset
Idepix
algorithm
blending
Chlorophyll-a
Suspended
matter
Trophic state
classes
Turbidity
10-d
aggregate
Radiometric
correction
Representative
spectrum
Copernicus Global
Land Service –
Lake Water v1.1.0
Architecture and performance
10. Performance: GloboLakes
10-yr time-series for 1000 lakes
3 – 45 minutes
per satellite pass per lake
[or: 5 min avg per lake = 3.5 CPU days]
Processing grid: 800 CPU cores
Processing time: 7y 51d 2h 24m
Input data volume (global archive): 136 TB
Output data volume: 11.4 TB
11. Making data visible: GloboLakes portal
screenshot of globolakes.eofrom.space of monthly average chlorophyll-a (300m), July 2011
14. Monthly time series over box in northern Tanganyika
ESA Ocean Colour CCI Portal
15. Data at 4-km but 1-km daily
also demonstrated
ESA Ocean Colour CCI Portal
16. State-of-the-art
• Large and mostly clear open waters observable since 1997
• MERIS sensor (2002-2012) kicked things off for inland waters
• Gradually moving from regional success stories to global services:
– Lake-specific algorithms often outperform the water-type specific global algorithms..
– but global approach gives us decent picture of remote or under-sampled sites
• Time-series data available on request, independent validation encouraged.
• We can add additional lakes to our processing chain & visualisation
• Community in situ data contributions (LIMNADES) essential for globally valid
result for all water types (we are not there yet).
17. Validation, uncertainties, anomalies, drift
In situ validation is a particular challenge in inland waters
- Wide optical diversity that changes every day, no climatologies or
‘ocean provinces’ to fall back on
- Moving towards: connectivity with high frequency in situ sensors,
citizen observatories, supersites, national monitoring databases,
LIMNADES, big data analytics
18. Improvements with
Sentinel-3 Ocean Land Colour Instrument
• Sentinel 3A launched in 2016, 20-yr mission
• Overlapping sensors from 2018, revisit times halved,
daily coverage (4 x data volume)
• Added wavebands:
– Improved atmospheric correction (under development)
– Improved substance retrieval in optically complex waters
• Sensor similar to MERIS so we can port algorithms over
• However, independent validation must continue
19. Sentinel-2 MultiSpectral Imager
Improvements
• Very high global resolution: 10, 20, 60 m bands
Challenges
• Mixed land/water optical configuration
– not optimal to retrieve optical water quality parameters
• Algorithm development and validation will take time
20. Conclusions
• Global inland water quality remote sensing requires specific data processing
• Retrieval of water constituent concentrations relies on continuous broad-
scale validation -> community effort
• New generation of satellites and related services (Copernicus) can cause a
step change in data-driven water quality management…
• Inland water remote sensing research needs to speed up to stay ahead of
sensor developments, or risk forever working with land/ocean sensors
• Algorithm research to continue, but hopefully with better focus on under-
sampled water types: healthy inland waters are relatively under-sampled by
optical water quality researchers -> focussing of effort needed.
Chicken and egg problem: success stories needed at regional level to influence
global satellite data uptake, algorithm development, and influence water quality
management (through policy). For successful and sustainable regional
application, extend support for in situ sampling and integrative research