Chlorophyll-a mapping from satellite remote-sensing after atmospheric correction and in situ validation (Sentinel-2, -3)

Prepared by Javier Maldonado Alfaro (IIGEO/UMSA)

& Xavier Lazzaro (BOREA/IRD)

Before developing a satellite algorithm, it is essential to carry out an atmospheric correction

It is essential to perform an atmospheric correction because it can significantly improve the quality of the remote sensing data by removing or compensating for atmospheric effects such as scattering, absorption, and reflection. They are the effects of atmospheric gases and particles on the light received by the sensor. They naturaly occur within the layer between the water surface and the satellite's optical sensor, and thus alter the water colour perceived by the satellite.  For what a correction is required (Moses et al. 2017). Atmospheric correction is more challenging for inland waters than for open ocean waters due to a number of factors. As inland waters are surrounded by various terrestrial sources of atmospheric pollution, the atmosphere results more optically heterogeneous. Therefore, the signal received at the sensor is 'contaminated'. This is more problematic when raised topography (mountains) surround the water body. That is the case of Lake Titicaca on the Altiplano, surrounded by the two Andes Cordilleras. 

There are several atmospheric correction algorithms available, and code implemented in softwares of image processing, but there is no consensus about which one should be used for remote sensing of the lakes water color. In our study, we compared the performances of the C2RCC, iCOR, ACOLITE, and Sen2Cor atmospheric corrections. The most suitable method is chosen by reducing the atmospheric interference, i.e.  matching best the corrected image remote sensing reflec- tance with the ‘truth ground’, which is is the water reflectance near the lake surface. To measure surface reflectance we used a spectroradiometer ASD HandHeld Pro 2 (Boulder, CO, USA), abbreviated to HH2, with the ViewSpec Pro Version 6.2.0 software (see the OLT book, pages 104-117). 

To validate the atmospheric correction, we conducted a total of 18 surface reflectance data collection daily campaigns, using the HH2 spectroradiometer, between November 27, 2018 and December 18, 2019. These in situ data are called "concordance data" because they are measured on the same day synchronously with the passage of the satellite Sentinel-2 (10:47), between 09:00 and 17:00, at different limnological stations of the initial network (41 stations in total; see the tab "Limnological stations of water quality). 

References

Moses W.J., Sterckx S. & Montes M.J.  (2017). Atmospheric correction for inland waters. In: Bio-Optical Modelling and Remote Sensingof Inland Waters, 1, 101-105. doi: 10.5268/iw-1.2.359. 

The use of in situ chlorophyll-a measurements to validate an algorithm for the images of the Sentinel-2 satellite

by Javier Maldonado Alfaro (IIGEO/UMSA)

& Xavier Lazzaro (BOREA/IRD)

• The ESA Sentinel-2: https://www.esa.int/Applications/Observing_the_Earth/Copernicus/Sentinel-2 

• Earth from Space: Lake Titicaca: https://www.esa.int/ESA_Multimedia/Keywords/Description/Lakes/(result_type)/videos_p

• Chlorophyll map of Lake Titicaca: https://www.esa.int/esearch?q=Titicaca  

• Top 10 Earth observation stories of 2023: https://www.esa.int/Applications/Observing_the_Earth/Top_10_Earth_observation_stories_of_2023

• ESA releases data to reveal how climate change impacts lakes: https://www.esa.int/Space_in_Member_States/United_Kingdom/ESA_releases_data_to_reveal_how_climate_change_impacts_lakes  

Is there a universal chlorophyll-a algorithm for Sentinel-2 images (or other satellite missions) in lakes?

Chlorophyll-a concentration is a major bioindicator of phytoplankton biomass, which further reflects the water quality and state of lakes. Various environmental factors may affect chlorophyll-a, such as phytoplankton composition, nutrients concentration (N, P), latitude, precipitation regime, altitude, temperature range, hydro-optical properties, colored dissolved organic matter (CDOM), inorganic suspended minerals (SM), among others. Lakes have more optically complex water than oceans. The algorithms used by NASA for the open oceans are empirical, visible (blue/green) band ratio techniques are effective as they are dominated by phytoplankton, and thus referred to as Case I waters by Morel & Prieur (1977). In contrast, the waters of lakes and coastal ocean areas are known as the Morel & Prieur's Case II waters. Because of the diverse characteristics of lake waters, it is necessary to develop chlorophyll-a algorithms applicable to satellite imagery, specific to each lake based on in situ monitoring databases. They result in more accurate predictions of chlorophyll-a concentrations, and thus contribute to anticipate eutrophication, then to prevent, control, and treat algal blooms. Shuchman et al. (2006) demonstrated that the combination of chlorophyll-a (Chl-a), CDOM, and SM generate the color of the lakes observed from space.

References

Morel A. & Prieur L. (1977). Analysis of variations in ocean color. Limnology and Oceanography, 22(4), 709-722. https://doi.org/10.4319/lo.1977.22.4.0709 

Shuchman R., Korosov A., Hatt C., Pozdnyakov D., Means J. & Meadows M. (2006). Verification and Application of a bio-optical algorithm for Lake Michigan using SeaWiFS: a 7-year inter-annual analysis. Journal of Great Lakes Research, 32(2), 258-279. 

Sentinel-3 and the Optical Water Type (OWT)

The range of spectral absorption and scattering between and within global lakes is extremely variable. Yet, satellite optical sensors have the advantage to capture the distinctly different colors of lakes, registering their Apparent Optical Properties (AOPs). Thus, satellites can identify and map water types with distinctly different reflectances and total Inherent Optical Properties (IOPs). Total IOPs depend on absorption and scattering of pure water and optically active constituents dissolved and suspended in the water column. The main optically active substances are chlorophyll-a (Chl-a; i.. the man photosynthetic pigment of phytoplankton), colored dissolved organic matter (CDOM), total suspended matter (TSM), and occasionally phycocyanin (PC; i.e. the photosynthetic pigment of cyanobacteria; contribution poorly studied). The high overpass frequency of Sentinel-3 (S-3, daily) and the high spatial resolution of Sentinel-2 (S-2, 10 m) provide unprecedented monitoring capabilities of lake surfaces. 

To map the trophic state of Minor Lake, Javier A. Maldonado A. (IIGEO/UMSA; unpublished) used the Optical Water Type (OWT; Moore et al., 2014) classification applied on Sentinel-3 images. This classification is based on 6 class ranges combining the median, minimum and maximum of 3 key indicators of water quality: Chlorophyll-a (Chl-a); Colored Dissolved Organic Matter (CDOM) that refers to organic matter in water that absorbs strongly in the ultraviolet (UV) spectrum; and Total Suspended Matter (TSM) that affects the water optical properties through absorbing and scattering the sunlight and can influence the photosynthetic process and total primary production of phytoplankton and macrophytes (see Table below).  Class 4 (deep blue) is the most eutrophic with 107 µg/L of median Chl-a concentration (min = 0.9 µg/L; max = 705 µg/L) that can reflect phytoplankton efflorescence (bloom). This OWT algorithm has been designed as an optical pre-classification algorithm for main water-types of lakes by the European Space Agency (ESA) for both Sentinel-2 and Sentinel-3 images through the Global Lakes Sentinel Services (GLaSS) grant number 313256 (Eleveld et al., 2014). 

References

Eleveld M.A., Ruescas A. & Hommersom A. (2014). WP3 Algorithm development for S2 and S3 (pre-processing and data reduction) - D3.3 Optical pre-classification method. Global Lakes Sentinel Services, ESA Grant number 313256, 84 pp.

Moore, T.S., Dowell, M.D., Bradt, S., Ruiz Verdu, A. (2014). An optical water type framework for selecting and blending retrievals from bio-optical algorithms in lakes and coastal waters. Remote Sensing of Environment 113(11), 2424–2430.

Table 1 - Classification GLaSS_6C of ESA: characteristics and ranges of the 6 OWT Classes. Chl-a in µg/L; CDOM in RFU (Relative Fluorescence Unit); TSM in g/m3. 

Figure 1 - The image carousel above shows a sequence of S-3 OWT maps from October 22 to December, 2003. It can be used as a timelapse, to visualize the high spatial-temporal dynamics of the eutrophication process in Minor Lake, and also as  an early warning of phytoplankton efflorescence (bloom). As such, the Class 4 (deep blue) spread in most of the northern and central regions on 16/11, and 09/12 (reaching 32% of the overall area of Minor Lake) as shown by the highest Chl-a concentrations registered by the YSI EXO2 probe of the HydroMet buoy (see the <Buoy water quality 2023-2024> thumbnail. The most eutrophicated area spreads from Cojata island (facing Huarina) to the littoral of Cohana bay down to the northern shore of Taraco peninsula. Indeed, the wastewaters of Batallas river flow into the lake in front of Cojata island, and of the Katari watershed into Cohana Bay, initially close to the Cohana community and presently from the Chojasivi community. 

Determining the 1979-2019 evolution of the Totora coastal extension in Lago Menor using Landsat-2, 5 and 8 imagery

by Jhasmin Duarte Tejerina & Mishel Justiniano Ayllón (IIGEO/UMSA, La Paz) 

Totora (Schoenoplectus californicus) is an emergent aquatic macrophyte, endemic to the Altiplano. Its role as natural biofilter of excess nutrients, organic matter, and certain (metalic) contaminants is well established, as well as its role in biodiversity conservation for fish, frogs, and waterfowl, in particular. 

The perception of the evolution of the spatial-temporal extension of Totora is very controversial among the coastal populations. The negative impacts of the increasing contamination through time, as well as the role of the water level fluctuation,  have frequently been overstated. Yet, reliable quantitative observations and experiments are still mostly missing. 

For these reasons, we got interested in elucidating this question, and thus implemented a monitoring over the longest period possible, that is the last four decades during which satellite imagery have been available. 

To calculate the area of Totora reeds distributed in the Titicaca Lago Menor, the years with extreme events of the last four decades (1979, 1986, 1996, 1996, 1997, 2004, 2010, 2013, 2015) were considered, in addition to the two years in which field trips were made for this study (2018 and 2019) (Fig. 2). Subsequently, with the help of Landsat-2, 5 and 8 satellite images downloaded from the United States Geological Survey (USGS) platform, we performed a supervised classification for each year of the study, taking into account the dry and wet seasons.

Supervised classification - Based on the campaigns carried out in the study area and a previous reconnaissance of the terrain, Landsat-2, 5 and 8 satellite images were downloaded with the help of the ENVI 5.3 software. To begin the supervised classification, an image is displayed with a combination of bands (6, 5, 4) that allows the analysis of the vegetation, differentiating the existing coverages in the image. Then, regions of interest ('ROI') were taken and a name was assigned to each sample taken. Samples were taken from rocky cliffs, clouds, water, rivers, vegetation, glaciers, crops, urban area and totora. A total of 10 regions of interest were sampled. We proceeded to perform the analysis of separability of training areas so that the different coverages to be classified is sufficiently independent and distinguishable from each other. Next, the raster obtained was converted into vector data, so that the totora can be separated from other coverages and the area in hectares of different images can be calculated (Fig. 3). 

Reflectance and spectral signatures - We used the ASD FieldSpec 600860 Handheld Spectroradiometer 2 (ASD Inc., Boulder, CO 80301 USA) [https://www.academia.edu/35089421/ASD_Document_600860_Rev._A_1_FieldSpec_HandHeld_2_User_Manual_FieldSpec_HandHeld_2_Spectroradiometer_Users_Manual] to measure the reflectance by obtaining spectral signatures of water and aquatic macrophytes, among other covers, in the 325-1,075 nm range [http://www.geo-informatie.nl/courses/grs60312/material2017/manuals/600860-dHH2Manual.pdf]. This instrument (ASD 2010) was previously calibrated with a reference target at each sampling station. We collected 5 to 10 spectral signatures from totora, at a vertical distance of 30 cm, between 10:00 and 16:00 on the day and around the 10:47 time of the Sentinel-2 satellite pass. This schedule presents the optimal conditions for the capture of spectral signatures as long as there is the least amount of clouds that prevent the passage of sunlight and cause a bad shot, with noise in the signatures. For the spectral signatures capture, we programmed the field trips with the passages of Landsat-8 (14:18, frequency 18 days) and Sentinel-2 (10:47, frequency 5 days) satellites. Because of this difference in passage frequencies, Landsat-8 and Sentinel-2 rarely pass on the same day. A healthy plant has higher reflectance values than one that is senescent or in poor condition. Thus, in the signatures taken in the field, the totora in good condition (Fig. 4-b) has a high reflectivity relative to the signature of the senescent cattail (Fig.4-c). The signatures per group were averaged according to condition (young or senescent). As a result, the reflectance values of each band were obtained in ASCII format. These values can be used to classify the phenological state during the dry and wet seasons in 2019.

Quadrat method - The quadrat method consists of placing a square (here a 1 m x 1 m PVC tubing) over the totora to determine the density, coverage and frequency of these plants (Fig. 5). A database was created in which, the first column contains the recorded number of totora/m2 at each station, and in the second column the NDVI value* (Landsat Normalized Difference Vegetation Index), which was extracted from the raster generated previously according to the location of each study station. In order to determine the relationship between stem density/m2 and NDVI, we performed a Pearson's correlation, which resulted as highly significant: R = 0.859 (Fig. 6). To estimate the totora density during 2019, we made a calculation with the different tools of ArcGIS 10.6, using the quadratic equation and the NDVI raster generated with Sentinel-2A images, both during the dry and wet seasons, obtaining a 'shape file' with the density values.

*NDVI is delivered as a single band product: in Landsat 4-7, NDVI = (Band 4 – Band 3) / (Band 4 + Band 3), and in Landsat 8-9, NDVI = (Band 5 – Band 4) / (Band 5 + Band 4).