Day 4

Detailed paper information

Back to list

Paper title Lakes’ geometrical and bio-optical properties determine the feasibility to derive water quality information using Sentinel-2 MSI data
Authors
  1. Nikolay Moshenskiy Tartu Observatory, University of Tartu, Estonia Speaker
  2. Krista Alikas Tartu Observatory, University of Tartu, Estonia
  3. Kersti Kangro Tartu Observatory, University of Tartu
Form of presentation Poster
Topics
  • A7. Hydrology and Water Cycle
    • A7.06 EO for monitoring water quality and ecological status in inland waters
Abstract text Monitoring is an integral precondition to determine lakes' ecological status and develop solutions to restore lakes that have deteriorated from reference conditions. Spatial and temporal limitations of conventional in situ monitoring impede adequate evaluation of lakes' ecological status, especially when dealing with large-scale measurements. Sentinel-2 (S2) - a constellation of the two twin satellites, S2-A and S2-B with the MultiSpectral Instrument (MSI) on board can be a complement to in situ data. S2 MSI imagery makes possible investigation of even small water bodies due to its high spatial resolution of 10, 20 and 60 meters depending on the spectral band. Besides, S2 spectral resolution allows estimation of a wide range of water quality parameters such as chlorophyll-a (chl-a), water color, colored dissolved organic matter (CDOM), etc. However, implementing the remote sensing data for water quality assessment over small inland waters might be obstructed by the adjacency effect (AE). AE is especially strong in small, narrow, or complex-shape water bodies surrounded by dense vegetation and decreases further offshore. Therefore, the largest possible homogeneous water area surrounding the sampling point would increase the possibility to obtain an accurate signal from the water’s surface called water-leaving reflectance ρω(λ). Moreover, a combination of chl-a, CDOM and TSM concentrations also affect the probability and accuracy of ρω(λ) and must be considered.
Test sites of this study are optically complex lakes of Northern Europe with a high and varying amount of optically active substances. In this study, the dataset of 476 in situ measurements of water properties from 44 lakes were used. Measured concentrations of chl-a are ranged between 2 mg/m3-100 mg/m3, total suspended matter (TSM) 0.6 mg/m3 - 48 mg/m3 and aCDOM (442) 0.5 – 48 m-1. Water-leaving reflectance ρω(λ) was measured deploying above-water RAMSES TriOS radiometers.
The aim of this study was to evaluate the capabilities and limitations of the S2 MSI data after atmospheric correction by POLYMER 4.12 and C2RCC v1.5 processors. The results were analysed together with lakes’ area and shape complexity (shape index, SI) and the signal strength as determined by the concentration of chl-a, TSM and aCDOM.
The objectives of the study were:
1. Validate and analyze POLYMER and C2RCC-derived ρω(λ) against in situ measurements using match-up analysis for exact location (1 x 1), 3 x 3 and 5 x 5-pixel size region of interest (ROI).
2. Evaluate spatial distribution and homogeneity of POLYMER and C2RCC quality flags and water quality products. Based on that derive area and SI thresholds of the lakes that can be monitored with S2 (20 m spatial resolution).
3. Evaluate the spatial and temporal distribution of the failures in POLYMER and C2RCC atmospheric correction and in the resulting water quality maps. Analyze its impact on the derived ecological status class in optically different lakes.
The validation of POLYMER ρω(λ) product against in situ measurements resulted in slightly better accuracy than the C2RCC product. For the bands at 560 nm, 665 nm and 705 nm, crucial to derive chl-a over optically complex waters, POLYMER showed a weak correlation (R2 = 0.41, 0.12, 0.36) for 1 x 1 area, however, R2 for the 3 x 3 region was higher and equaled 0.63, 0.48 and 0.58, respectively. Noticeably, with enlarging the ROI up to 5 x 5 pixels grid, R2 decreased and equaled 0.36, 0.33 and 0.31 for 560 nm, 665 nm and 705 nm, respectively, which indicates nonhomogeneity in pixels distribution. Moreover, 5 x 5 ROI is 10000 m2 area, which might be too large to compare with the field measurement from only one point. The coefficient of determination for C2RCC data increased with the enlargement of the ROI to a 3 x 3 and 5 x 5-pixel area similar to POLYMER, however, not so noticeably. Specifically, for exact location (1 x 1 ROI), R2 equaled 0.45, 0.35, 0.40 at 560 nm, 665 nm and 705 nm wavebands, whereas for 3 x 3 and 5 x 5 pixel area it equaled 0.48, 0.41, 0.40 and 0.50, 0.38, 0.40, respectively.
POLYMER quality flags of the S2 imageries sensed in spring, summer and autumn over a group of 1727 lakes predominantly located in Southern Estonia were analyzed. In spring, most of the water bodies under 1 ha did not have valid quality flags. Besides, more complex-shape water bodies (SI > 2) with no valid quality flags were even larger (up to 6 ha). It was shown that the amount of quality flags, usable to produce water quality maps, decreases towards autumn.
Spatial and seasonal evaluation of the chl-a was conducted in the optically and geometrically different lakes. Failures in POLYMER atmospheric correction resulting in abnormal chl-a values were mostly due to the combined effect of optical properties of the water bodies and adjacency effect, strongest over clear waters surrounded by forest. This resulted in very few pixels and also with high spatial heterogeneity over clear water lakes. Whereas over eutrophic waters, there were more quality controlled satellite retrievals with improved spatial patterns of chl-a. It was shown that S2 MSI is a promising method for studying water bodies but the adjacency to the shore and the level of optically active substances must be considered.