Original ArticleTo what extent can Ulva and Sargassum be detected and separated in satellite imagery?
Introduction
Various types of floating macroalgae have been detected in satellite imagery in the world's oceans (Smetacek and Zingone, 2013; Thornber et al., 2017; Bermejo et al., 2019; Qi et al., 2020). Of particular importance are the recurrent blooms of Ulva prolifera and Sargassum horneri in the western Yellow Sea (YS) (Liu et al., 2009; Hu et al., 2010; Qi et al., 2016) and East China Sea (ECS) (Qi et al., 2017; Xing et al., 2017), respectively, as they have posed various environmental and economic problems as well as ecological consequences. These macroalgae have been studied extensively using a variety of field, laboratory, and remote sensing techniques (Liu et al., 2010; Liu et al., 2018; Kong et al., 2018; Byeon et al., 2019; Min et al., 2017, 2019), as they may be harmful to the coastal environments (e.g., oxygen consumption, beach pollution, insect attraction, bad smells, etc.) although these algae do not contain toxins. While each technique has its own strengths and weaknesses, satellite remote sensing can provide synoptic and frequent as well as historical data to study their distributions and temporal changes in order better understand their potential driving mechanisms and consequences (e.g., Qi et al., 2016, 2017; Xing et al., 2017, 2019; Zhang et al., 2020).
However, two inherent limitations exist in satellite remote sensing, which may bring uncertainties in estimating the algae distributions and areal coverage as well as tracing algae originations.
The first is the detection limit, i.e., at what sub-pixel coverage can these small features in satellite imagery be detected, as the algae patches are often scattered to cover a small portion of an image pixel. For example, using image statistics, Hu et al. (2017) showed that >99.5% of Ulva-containing image pixels (250-m resolution) do not have full pixel coverage, and > 90% of these pixels have sub-pixel Ulva coverage of < 5%. The sub-pixel coverage, χ (0.0% - 100%), represents the amount of algae (in m2) within a pixel of finite size, where the amount can be from many scattered patches within a pixel. Using simulations, Hu et al. (2015) showed that, under typical conditions, the detection limit (i.e., the smallest χ) only depends on the sensor's signal-to-noise ratios (SNRs) of the red and near-infrared (NIR) bands that are used to quantify the pixel's red-edge reflectance (in practice, through an Floating Algae Index or FAI, Hu, 2009). For example, with SNRs of 200:1, Hu et al. (2015) estimated a detection limit of χdet= 1% when FAI was used to detect floating vegetation. This estimate was later confirmed by Wang and Hu (2016) using data collected by the Moderate Resolution Imaging Spectroradiometer (MODIS) 1-km data – with SNRs of 1000:1 the subpixel detection limit from image statistics was estimated to be χdet=0.2%. So, the detection limit is inversely proportional to SNRs. Once below the detection limit, the floating vegetation within the pixel is difficult or impossible to detect. However, these are for typical clear atmosphere and clear waters (e.g., for Sargassum in the Atlantic Ocean). For turbid atmosphere and/or turbid waters, the detection limit may be different.
The second is the discrimination limit, i.e., above what sub-pixel coverage can these small features be differentiated to be a certain type of floating vegetation as opposed to others. Without a priori knowledge of the local environment, the discrimination can only rely on the spectral characteristics of the pixel. Hu et al. (2015) used simulations to show that in order to detect the chlorophyll-c absorption feature around 625 nm from Sargassum-containing pixels, a subpixel fraction of χ=25% is required for a sensor with SNRs of 200:1. Clearly, discriminating floating algae type is more demanding than detecting floating algae presence. Furthermore, depending on the algorithms used to discriminate algae type, the discrimination limit may also be different. Indeed, the 25% requirement is based on the assumption that no other information is available on the local environment and all types of floating matters (macroalgae, microalgae, weathered oil, dead seagrasses, garbage patches, etc.) may coexist in the same region. In reality, such situations are rare, and the discrimination is often between two types of floating algae for certain regions, for example between Sargassum (either S. flutains or S. natans) and Trichodesmium for the tropical Atlantic Ocean, or between Ulva and Sargassum horneri in the YS and ECS (Xiao et al., 2020). In such cases, discrimination between two types of floating algae may be much easier, with relaxed requirements on subpixel coverage. However, to date, such requirements have not been quantified. Specifically, to what extent can Ulva and Sargassum horneri be separated in satellite imagery?
The objective of this study is to address the two questions on the detection and discrimination limits of Ulva and Sargassum horneri using multi-sensor measurements, noise propagation theory, and simulations. The ultimate goal is to provide a general guide on the capacity of satellite sensors in mapping Ulva and Sargassum, while at the same time demonstrate an approach to quantify such capacity for other floating matters in other regions.
Throughout this study, the subpixel coverage (and therefore detection limit) is expressed as a percentage (χ from 0.0% to 100%) of the pixel size rather than a physical area (m2), although they are equivalent for any satellite sensor with a finite pixel size or spatial resolution. Once χ is known, the corresponding physical area is simply a product of χ and pixel size. In presenting algae distribution maps with gridded data, the algae concentration in each grid is also expressed in χ, representing algae density within the grid (Qi et al., 2016, 2017; Wang and Hu, 2016), equivalent to m2 (algae) per km2 (area). This is regardless of whether the algae is in a single patch or multiple patches within the pixel. When the field or laboratory determined calibration constant is available to convert algae area (m2) to biomass, χ can also be easily converted to algae biomass density (kg algae per pixel or per grid), as shown in Hu et al. (2017) and Wang et al. (2018). Although the calibration constant is not applicable above a certain “saturation” point (Fig. 6 of Hu et al., 2017), because algae density is way below the “saturation” point in most waters and this is especially true regarding the detection limit, the “saturation” has no impact on this work. For this reason, χ is used in this study to represent subpixel algae coverage or areal density.
Section snippets
Laboratory and field data
Reflectance data of Ulva and Sargassum horneri from water tank experiments were taken from Hu et al. (2017) and Huang et al. (2018), respectively (Fig. 1a). They are used in this study to represent typical spectral endmembers of floating macroalgae with 100% subpixel coverage (i.e., χ = 100%).
Reflectance of three water types over the YS and ECS were taken from Lee et al. (2016). They range from clear water, moderately turbid water, to turbid water, respectively (Fig. 1b). In this study, they
SNRs and data product noise
A sensor's sensitivity is often defined as its SNRs at typical radiance input (Ltyp, mW cm−2 μm−1 sr−1), where SNR of a specific band is
Here, sensor noise in the above equation, σ (mW cm−2 μm−1 sr−1), is defined as the standard deviation (i.e., a scalar quantity) of a univariate Gaussian distribution of random noise errors. This is similar to the definition of measurement uncertainties that is used to define the probability of a measured (or estimated) property in approaching the
Data product noise and uncertainty
Because σ(Rλ) depends on both water and atmosphere conditions, OLCI and MSI images collected over variable conditions were used to estimate σ(Rλ), specifically from relatively clear waters, moderately turbid waters, and highly turbid waters in the YS and ECS, and from clear and turbid atmospheres (aerosol optical thickness at 865 nm changed between ~0.06 and ~0.25). Fig. 4a and b shows some representative locations where σ(Rλ) was estimated. Their corresponding water-leaving reflectance (Rw = πR
Single band or band-combination index
On one hand, detection of floating algae presence is similar to detection of bright targets (e.g., oil/gas platforms (Liu et al., 2018) or ships (Heiselberg, 2016)) in the vast ocean, as they are all based on the concept of spatial anomalies. On the other hand, while the latter can be from single-band images (typically in the NIR) through the use of a roaming window and statistics-based threshold (Liu et al., 2018), the technique suffers from significant noise perturbations and lack of ability
Conclusions
Recurrent Ulva and Sargassum blooms in the Yellow Sea and East China Sea have stimulated research on their causes and consequences. While satellite remote sensing played a key role in monitoring and tracking the blooms, understanding their uncertainties requires accurate knowledge on their subpixel detection and discrimination limits. Using measurements from two popular satellite sensors, OLCI and MSI, and through simulated experiments, these limits are determined for most observing conditions.
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgments
This work was supported by the National Natural Science Foundation of China (No. 41806208, No. 42076180) (Qi). We thank the European Space Agency, the U.S. NASA, and the U.S. Geological Survey for providing all satellite data. We thank the two anonymous reviewers for their thorough comments to help improve the presentation of this work.
References (43)
- et al.
Spatial and temporal variability of biomass and composition of green tides in Ireland
Harmful Algae
(2019) A novel ocean color index to detect floating algae in the global oceans
Remote Sens. Environ.
(2009)- et al.
Spectral and spatial requirements of remote measurements of pelagic Sargassum macro algae
Remote Sens. Environ.
(2015) - et al.
Remote estimation of biomass of Ulva prolifera macroalgae in the Yellow Sea
Remote Sens. Environ.
(2017) - et al.
World's largest macroalgal bloom caused by expansion of seaweed aquaculture in China
Mar. Pollut. Bull.
(2009) - et al.
Recurrence of the world's largest green-tide in 2009 in Yellow Sea, China: porphyra yezoensis aquaculture rafts confirmed as nursery for macroalgal blooms
Mar. Pollut. Bull.
(2010) - et al.
Assessment of offshore oil/gas platform status in the northern Gulf of Mexico using multi-source satellite time-series images
Remote Sens. Environ.
(2018) - et al.
Airborne remote sensing of macrophytes in Cefni Reservoir, Anglesey, UK
Aquat. Bot.
(1997) - et al.
Long-term trend of Ulva prolifera blooms in the western Yellow Sea
Harmful Algae
(2016) - et al.
In search of floating algae and other organisms in global oceans and lakes
Remote Sens. Environ.
(2020)
Spatial and temporal variability in macroalgal blooms in a eutrophied coastal estuary
Harmful Algae
Advantages of high quality SWIR bands for ocean colour processing: examples from Landsat-8
Remote Sens. Environ.
Remote sensing estimation of the biomass of floating Ulva prolifera and analysis of the main factors driving the interannual variability of the biomass in the Yellow Sea
Mar. Poll. Bull.
An anomalous bi-macroalgal bloom caused by Ulva and Sargassum seaweeds during spring to summer of 2017 in the western Yellow Sea, China
Harmful Alage
Monitoring seaweed aquaculture in the Yellow Sea with multiple sensors for managing the disaster of macroalgal blooms
Remote Sens. Environ.
Characterization of the development stages and roles of nutrients and other environmental factors in green tides in the Southern Yellow Sea, China
Harmful Algae
Phenology of Trichodesmium spp. blooms in the Great Barrier Reef lagoon, Australia, from the ESA-MERIS 10-year mission
PLoS ONE
The origin and population genetic structure of the ‘golden tide’ seaweeds, Sargassum horneri, in Korean waters
Sci. Rep.
Spatial-temporal distribution of golden tide based on high-resolution satellite remote sensing in the South Yellow Sea
J. Coast. Res.
Atmospheric correction of ocean color imagery in the earth observing system era
J. Geophys. Res.
Ocean color satellites show extensive lines of floating Sargassum in the Gulf of Mexico
IEEE Trans. Geosci. Remote Sens.
Cited by (39)
Detecting Marine pollutants and Sea Surface features with Deep learning in Sentinel-2 imagery
2024, ISPRS Journal of Photogrammetry and Remote SensingCurrent status and prospects of algal bloom early warning technologies: A Review
2024, Journal of Environmental ManagementMapping and quantifying pelagic Sargassum in the Atlantic Ocean using multi-band medium-resolution satellite data and deep learning
2023, Remote Sensing of EnvironmentMapping Ulva prolifera green tides from space: A revisit on algorithm design and data products
2023, International Journal of Applied Earth Observation and Geoinformation