Remote Sensing techniques for inland waters are spreading as a useful, low cost, high frequency, auxiliary tool for water quality monitoring, supporting essential but more rare and punctual limnologic in situ measurements. In particular, the launch of the latest generation Sentinels sensors, provided products characterised by high frequency, medium-high resolution, high signal-to-noise ratio, long time series, thanks to the future planned launches of twin satellites. Especially Sentinel-3 OLCI, as successor of MERIS, with unique band settings and a 300 m spatial resolution has been offering valuable data for observing almost daily water quality parameters in multiple medium-large water basins at global scale. On the other hand, to be used for operational monitoring, water quality parameters products have to be mapped with high accuracy which, as preliminary step, requires accurate atmospheric correction, e.g. the removal from the signal of the contribution due to the presence of the atmosphere and other disturbances such as effects due to water interface and adjacency of lands. In inland and transitional systems, where optically complex waters dominate due to the different origins of water constituents (due to e.g. coastal discharges and runoff, column overturn, resuspension of bottom sediments), the atmospheric correction might be particularly tricky. In particular, residual backscattering in the NIR region due to suspended matter prevents from using traditional algorithms based on the NIR complete absorption assumption, used over clear waters; secondly, the presence of the close land as well as of a severe topography might cause additional contribution to the signal (adjacency effect), which should not be ignored but removed prior to run the algorithms to retrieve for water quality parameters. The aim of this work is to evaluate the accuracy of Sentinel3A-OLCI Level-2 data products in three inland waters basins, with increasing eutrophic state, from the deep oligotrophic subalpine Lake Garda, in North Italy, to the shallower, more eutrophic and turbid Lake Trasimeno (Centre Italy) and the iper-eutrophic Curonian Lagoon (Lithuania-Russian). Chlorophyll (chl-a) concentration from both traditional and Neural Network (NN) products were validated, being compared to available in situ measurements. In particular, 12 matchup points over Lake Garda, 8 over Lake Trasimeno, and 16 over Curonian Lagoon were exploited to evaluate the performance of these products. Where in situ radiometric measurements were available, Remote Sensing Reflectance (Rrs) retrieved by standard atmospheric correction and from NN (estimated from Level-1B products, through SNAP-C2RCC OLCI tool) were validated. Spectral angle (SAM) and chi-squared were calculated for each measurement in order to evaluate performances not only in terms of magnitude but also of the overall shape of the estimated spectrum. Both Rrs and chl-a concentration products were compared to the results obtained through an alternative processing chain. A 6SV-based atmospheric correction approach, parametrised according to AERONET aerosol parameters (if available) or to MODIS-retrieved products, was used over L1B images. Rrs obtained were used to estimate chl-a concentration through the bio-optical model BOMBER (Bio-Optical Model Based tool for Estimating water quality and bottom properties from Remote sensing images, Giardino et al., 2012) for Lake Garda and Lake Trasimeno and an empirical algorithm based on band ratio (Bresciani et al., 2016) for Curonian Lagoon. In addition, over Lake Garda, Aerosol Optical Thickness (AOT) product was compared to AOT estimates, from the AERONET station located in Sirmione (on the Southern coast of the Lake), looking for any correlation between AOT estimate error and chl-a estimate error. Finally, for the two Italian lakes, chl-a concentration estimates were extracted over the in situ stations used by regional environmental protection agencies for water quality monitoring and WFD activities. Using suggested flags for NN products extraction, from April 2016 to July 2017, 161 cloud and high glint risk free images were used for Lake Garda and 144 for Lake Trasimeno in order to evaluate seasonal trends, compared to in situ and selected 6SV-BOMBER products. Concerning Rrs results over Lake Garda, considering the subset of bands used through the BIO-OPTICAL model to retrieve chl-a concentration (bands 2 to 10), statistics showed comparable results for 6SV-based algorithm results and NN results: SAM was 15.9, 49.2, and 15.7 degrees and chi-squared was 0.040, 0.055, and 0.044 for 6SV-based, Level2 standard and NN algorithms respectively. On the other hand, plotting estimated spectra against in situ measurements, it is still evident how NN products tends to highly underestimate values at 560 nm, essential for chl-a retrieval. Over Lake Trasimeno, only 2 dates for a total of 8 matchups were available: first results showed a better performance by NN algorithm with chi-squared equal to 0.0098, while it was 0.038 and 0.29 for 6SV-based and standard products respectively. Finally over Curonian Lagoon, 6SV-based and NN products were the best performing in terms of SAM (10 and 17 degrees respectively). Considering band ratio used by semi-empirical algorithm, over Curonian Lagoon the best estimates were obtained for 6SV-based products: MAPE mean absolute percentage error was 8% for 6SV-based, 19% for Level2 standard and 36% for NN algorithms respectively. On the other hand, MAPE obtained for Lake Garda chl-a concentration was 39%, 153%, and 86% for respectively BOMBER, standard (CHL_OC4ME) and NN chl products (CHL_NN). Over Lake Trasimeno MAPE was 23% for BOMBER and 56 % for NN. Results showed that over inland waters atmospheric correction requires more investigation, especially for the most oligotrophic waters, where the concentration of TSM and Phytoplankton is particularly low. Even if statistics showed similar results for atmospheric correction performances, the best results in terms of chl-a concentration were obtained through the BIO-OPTICAL or semi-empirical, sites-specific algorithms. New data will be added to enlarge this dataset as they come available through reprocessing and new field campaigns.

Cazzaniga, I., Bresciani, M., Giardino, C., Vaiciute, D. (2018). Qualification of atmospheric correction and water quality retrieval from OLCI for European inland and transitional waters. Intervento presentato a: 4th S3VT meeting, Darmstadt - Germania.

Qualification of atmospheric correction and water quality retrieval from OLCI for European inland and transitional waters

Cazzaniga, I;
2018

Abstract

Remote Sensing techniques for inland waters are spreading as a useful, low cost, high frequency, auxiliary tool for water quality monitoring, supporting essential but more rare and punctual limnologic in situ measurements. In particular, the launch of the latest generation Sentinels sensors, provided products characterised by high frequency, medium-high resolution, high signal-to-noise ratio, long time series, thanks to the future planned launches of twin satellites. Especially Sentinel-3 OLCI, as successor of MERIS, with unique band settings and a 300 m spatial resolution has been offering valuable data for observing almost daily water quality parameters in multiple medium-large water basins at global scale. On the other hand, to be used for operational monitoring, water quality parameters products have to be mapped with high accuracy which, as preliminary step, requires accurate atmospheric correction, e.g. the removal from the signal of the contribution due to the presence of the atmosphere and other disturbances such as effects due to water interface and adjacency of lands. In inland and transitional systems, where optically complex waters dominate due to the different origins of water constituents (due to e.g. coastal discharges and runoff, column overturn, resuspension of bottom sediments), the atmospheric correction might be particularly tricky. In particular, residual backscattering in the NIR region due to suspended matter prevents from using traditional algorithms based on the NIR complete absorption assumption, used over clear waters; secondly, the presence of the close land as well as of a severe topography might cause additional contribution to the signal (adjacency effect), which should not be ignored but removed prior to run the algorithms to retrieve for water quality parameters. The aim of this work is to evaluate the accuracy of Sentinel3A-OLCI Level-2 data products in three inland waters basins, with increasing eutrophic state, from the deep oligotrophic subalpine Lake Garda, in North Italy, to the shallower, more eutrophic and turbid Lake Trasimeno (Centre Italy) and the iper-eutrophic Curonian Lagoon (Lithuania-Russian). Chlorophyll (chl-a) concentration from both traditional and Neural Network (NN) products were validated, being compared to available in situ measurements. In particular, 12 matchup points over Lake Garda, 8 over Lake Trasimeno, and 16 over Curonian Lagoon were exploited to evaluate the performance of these products. Where in situ radiometric measurements were available, Remote Sensing Reflectance (Rrs) retrieved by standard atmospheric correction and from NN (estimated from Level-1B products, through SNAP-C2RCC OLCI tool) were validated. Spectral angle (SAM) and chi-squared were calculated for each measurement in order to evaluate performances not only in terms of magnitude but also of the overall shape of the estimated spectrum. Both Rrs and chl-a concentration products were compared to the results obtained through an alternative processing chain. A 6SV-based atmospheric correction approach, parametrised according to AERONET aerosol parameters (if available) or to MODIS-retrieved products, was used over L1B images. Rrs obtained were used to estimate chl-a concentration through the bio-optical model BOMBER (Bio-Optical Model Based tool for Estimating water quality and bottom properties from Remote sensing images, Giardino et al., 2012) for Lake Garda and Lake Trasimeno and an empirical algorithm based on band ratio (Bresciani et al., 2016) for Curonian Lagoon. In addition, over Lake Garda, Aerosol Optical Thickness (AOT) product was compared to AOT estimates, from the AERONET station located in Sirmione (on the Southern coast of the Lake), looking for any correlation between AOT estimate error and chl-a estimate error. Finally, for the two Italian lakes, chl-a concentration estimates were extracted over the in situ stations used by regional environmental protection agencies for water quality monitoring and WFD activities. Using suggested flags for NN products extraction, from April 2016 to July 2017, 161 cloud and high glint risk free images were used for Lake Garda and 144 for Lake Trasimeno in order to evaluate seasonal trends, compared to in situ and selected 6SV-BOMBER products. Concerning Rrs results over Lake Garda, considering the subset of bands used through the BIO-OPTICAL model to retrieve chl-a concentration (bands 2 to 10), statistics showed comparable results for 6SV-based algorithm results and NN results: SAM was 15.9, 49.2, and 15.7 degrees and chi-squared was 0.040, 0.055, and 0.044 for 6SV-based, Level2 standard and NN algorithms respectively. On the other hand, plotting estimated spectra against in situ measurements, it is still evident how NN products tends to highly underestimate values at 560 nm, essential for chl-a retrieval. Over Lake Trasimeno, only 2 dates for a total of 8 matchups were available: first results showed a better performance by NN algorithm with chi-squared equal to 0.0098, while it was 0.038 and 0.29 for 6SV-based and standard products respectively. Finally over Curonian Lagoon, 6SV-based and NN products were the best performing in terms of SAM (10 and 17 degrees respectively). Considering band ratio used by semi-empirical algorithm, over Curonian Lagoon the best estimates were obtained for 6SV-based products: MAPE mean absolute percentage error was 8% for 6SV-based, 19% for Level2 standard and 36% for NN algorithms respectively. On the other hand, MAPE obtained for Lake Garda chl-a concentration was 39%, 153%, and 86% for respectively BOMBER, standard (CHL_OC4ME) and NN chl products (CHL_NN). Over Lake Trasimeno MAPE was 23% for BOMBER and 56 % for NN. Results showed that over inland waters atmospheric correction requires more investigation, especially for the most oligotrophic waters, where the concentration of TSM and Phytoplankton is particularly low. Even if statistics showed similar results for atmospheric correction performances, the best results in terms of chl-a concentration were obtained through the BIO-OPTICAL or semi-empirical, sites-specific algorithms. New data will be added to enlarge this dataset as they come available through reprocessing and new field campaigns.
poster
Reote Sensing, Inland waters, phytoplankton,
English
4th S3VT meeting
2018
2018
none
Cazzaniga, I., Bresciani, M., Giardino, C., Vaiciute, D. (2018). Qualification of atmospheric correction and water quality retrieval from OLCI for European inland and transitional waters. Intervento presentato a: 4th S3VT meeting, Darmstadt - Germania.
File in questo prodotto:
Non ci sono file associati a questo prodotto.

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/10281/204899
Citazioni
  • Scopus ND
  • ???jsp.display-item.citation.isi??? ND
Social impact