Detection of surface algal blooms using the newly developed algorithm surface algal bloom index (SABI) Fahad Alawadi*a aSchool of Ocean and Earth Science, University of Southampton, Waterfront Campus, Empress Dock, Southampton, S014 3ZH, UK ABSTRACT Quantifying ocean colour properties has evolved over the past two decades from being able to merely detect their biological activity to the ability to estimate chlorophyll concentration using optical satellite sensors like MODIS and MERIS. The production of chlorophyll spatial distribution maps is a good indicator of plankton biomass (primary production) and is useful for the tracing of oceanographic currents, jets and blooms, including harmful algal blooms (HABs). Depending on the type of HABs involved and the environmental conditions, if their concentration rises above a critical threshold, it can impact the flora and fauna of the aquatic habitat through the introduction of the so called “red tide” phenomenon. The estimation of chlorophyll concentration is derived from quantifying the spectral relationship between the blue and the green bands reflected from the water column. This spectral relationship is employed in the standard ocean colour chlorophyll-a (Chlor-a) product, but is incapable of detecting certain macro-algal species that float near to or at the water surface in the form of dense filaments or mats. The ability to accurately identify algal formations that sometimes appear as oil spill look-alikes in satellite imagery, contributes towards the reduction of false-positive incidents arising from oil spill monitoring operations. Such algal formations that occur in relatively high concentrations may experience, as in land vegetation, what is known as the “red-edge” effect. This phenomena occurs at the highest reflectance slope between the maximum absorption in the red due to the surrounding ocean water and the maximum reflectance in the infra-red due to the photosynthetic pigments present in the surface algae. A new algorithm termed the surface algal bloom index (SABI), has been proposed to delineate the spatial distributions of floating micro-algal species like for example cyanobacteria or exposed inter-tidal vegetation like seagrass. This algorithm was specifically modelled to adapt to the marine habitat through its inclusion of ocean-colour sensitive bands in a four-band ratio-based relationship. The algorithm has demonstrated high stability against various environmental conditions like aerosol and sun glint. Keyword List: surface algal bloom index, SABI, algae, MODIS, ROPME, HABs, NDVI, oil spill. 1. INTRODUCTION Oceanographers have studied satellite data in order to measure ocean color, from which they have been able to derive chlorophyll a (Chlor-a) concentration and phytoplankton biomass in open oceans (case I waters). This process started with the Coastal Zone Color Scanner (CZCS)1,2; and continued with a number of sensors: Ocean Color and Temperature Scanner–OCTS3; Sea-viewing Wide Field-of-view Sensor–SeaWiFS4; the Moderate Resolution Imaging Spectroradiometer–MODIS5; and the Medium Resolution Imaging Spectrometer – MERIS6. Chlorophyll causes a shift in the blue to green ratio of upwelling light in case I waters, where the optical properties mainly depend on phytoplankton7,8. The MODIS 500 m resolution Chlor-a algorithm was generated according to Equation (1): (1) where R is defined as: * [email protected]; [email protected] Remote Sensing of the Ocean, Sea Ice, and Large Water Regions 2010, edited by Charles R. Bostater Jr., Stelios P. Mertikas, Xavier Neyt, Miguel Velez-Reyes, Proc. of SPIE Vol. 7825, 782506 · © 2010 SPIE CCC code: 0277-786X/10/$18 · doi: 10.1117/12.862096 Proc. of SPIE Vol. 7825 782506-1 Downloaded from SPIE Digital Library on 22 Nov 2010 to 91.140.182.19. Terms of Use: http://spiedl.org/terms (2) where ρ is the remote sensing reflectance at the corresponding blue (B =469 nm) and green (B =555 nm) bands rs 3 4 respectively. Equation (2) has to yield a positive value to produce a valid Chlor-a pixel value, otherwise Chlor-a would fail to be evaluated for that particular pixel. This failure can be caused by the presence of absorbing aerosols that cannot be detected by the atmospheric correction process9, causing an overestimation of the atmospheric contribution in all bands. This overestimation will induce severe errors, particularly at the shorter wavelengths, that might even cause total failure in the production of Chlor-a10. Another reason for the failure of Chlor-a is infra-red (IR) reflectance caused by the presence of large concentrations of Chlor-a at the water surface or very close to it, which is a direct violation of the dark pixel assumption over clear water. Algae is a non-planktonic species that performs photosynthesis. It is commonly classified into two groups: macro-algae like seaweed and micro-algae like filamentous cyanobacteria. They are also divided into three groups based on their photosynthetic pigments: Phaeophyta (the browns), Rhodophyta (the reds) and Chlorophyta (the greens), but can also appear in various other colour combinations like yellow or silvery grey. They are capable of regulating their buoyancy to modulate their photosynthetic activity to the changing light levels. This mechanism of buoyancy is either to protect themselves from excessive radiation11 or to access more light from the photic region located in the top layer of the water surface11,12. Surface emerging algae can experience, similar to land-based vegetation, the “red-edge” effect, where the greatest reflectance slope occurs between the maximum absorption in the red and the maximum reflection in the infra- red (0.7 µm-1.0 µm). The IR reflectance is caused by the combination of decreasing absorption by the chlorophyll pigments in the spongy mesophyll cells and the increasing of absorption by water13-15. Texturally, surface blooms are often characterized by their thick extensive mats or scum that looks like oil slicks or slightly foamy pollution. Equation (3) shows the normalized difference vegetation index (NDVI) that was proposed by Rouse et al. (1974)16 to measure the vegetation activity at the land surface. It normalizes the output from –1 to +1, to partially account for the differences in illumination and surface slope. Researchers like Kahru et a1 (1993)17 have examined the use of NDVI on surface cyanobacterial blooms, but more recent algorithms were proposed specifically for the detection of surface algae like for example the Maximum Chlorophyll Index (MCI); the Spectral Contrast Shift (SCS); and the MODIS Floating Algae Index (FAI) proposed by Gower et al. (2005)18, Alawadi et al. (2008)19 and Hu (2009)20 respectively. (3) where X can be the top-of-atmosphere (TOA) radiance (L) – provided that it was measured in the same irradiance t conditions for both bands; TOA reflectance (ρ) or ρ at the near-infrared (NIR) and the red (R) bands. In the case of rs MODIS, the 250 m resolution bands are used at B (R =645 nm) and B (NIR=869 nm). 1 2 1.1 The Surface Algal Bloom Index (SABI) The SABI is an empirical algorithm developed in order to detect water floating biomass that has a NIR response similar to that of land vegetation, with the specific inclusion of ocean-sensitive spectral bands; the blue (characteristic of clear water) and green (characteristic of the water column bloom): (4) where X has been defined earlier for Equation (3), at the following MODIS wavelengths: B (X =645 nm) and B 1 R 2 (X =869 nm), both aggregated from the 250 m resolution MODIS band group; and bands B (X =469 nm) and B NIR 3 B 4 (X =555 nm), respectively available within the 500 m resolution MODIS band group. G The SABI formula is a ratio dependent relation, making it advantageous in cancelling out a large proportion of signal variations arising from the variations in the intensity of incident light that accompany different sun angles, Proc. of SPIE Vol. 7825 782506-2 Downloaded from SPIE Digital Library on 22 Nov 2010 to 91.140.182.19. Terms of Use: http://spiedl.org/terms clouds/shadow and atmospheric conditions21. It will however, not account for the spectral dependencies generated by the additive atmospheric (path radiance) effects. The inclusion of the blue and green bands in the denominator, and the subtraction of the red band in the numerator (Equation 4), are collectively expected to make the SABI less sensitive to atmospheric effects, particularly for those that are sensitive to the red and green bands. It is also expected to reduce the effect of molecular scattering (Rayleigh scattering) particularly at shorter wavelengths. 2. METHOD For this analysis, the relation for the TOA reflectance (ρ) that was used: (5) where L is the calibrated TOA radiance; F is the extraterrestrial mean solar irradiance, and θ is the solar zenith angle – t 0 the angle between the line from the pixel under examination to the sun and the local vertical. When substituting Equation (5) into Equation (4), the accuracy of the SABI will only depend on the uncertainties that exit in L and F . t 0 Although it is the ρ – produced after performing atmospheric correction, that is required as input in algorithms for rs deriving most of the MODIS ocean products22,23, the ρ was used instead. This is because performing atmospheric correction was deemed unnecessary at this preliminary phase, since the aim was to map the extent of algal blooms instead of producing a biophysical parameter like concentration24,25. Furthermore, the atmospheric correction assumes that case I type waters to have zero water-leaving radiance at the NIR bands (B =869 nm and B =748 nm for MODIS), 2 15 which conflicts with the red-edge effect exhibited by surface algae. The SeaWiFS Data Analysis System (SeaDAS) version 6.0 was used to process MODIS L1B TOA calibrated radiances to produce the ρ and other bio-optical products for correlation purposes like for example Chlor-a, Sea Surface Temperature (SST) and the Aerosol optical thickness at 869 nm (tau_869). Non-geometrically corrected images were also produced representing the natural RGB colour composite, corresponding to bands 1,4,3 (B =645 nm, B =555 nm, 1 4 B =469 nm) in the 500 m/pixel resolution, and a false RGB colour composite corresponding to the aggregated bands 3 2,1,1 in the 500 m/pixel resolution bands (B =645 nm, B =869 nm). The reason for composing these images was to 1 2 identify the natural colour of the bloom as it appears in the case of 143, and in the case of the 211 colour composite, to highlight the near-surface algae by appearing reddish in colour due to the NIR reflectance. The standard masks produced by SeaDAS during the atmospheric correction process like for example the cloud and sun glint masks were also used to test their masking quality over the valid surface algae pixels. 3. RESULTS Figure 1 shows the location map of the different datasets on which SABI was tested and the TOA spectral characteristics plot for the species involved, with respect to MODIS 500 m bands. 3.1 Sargassum– Gulf of Mexico Sargassum is characterized by having elongated, narrow fronds and numerous spherical gas-filled bladders called vesicles allowing the plant to float freely26. The colour of Sargassum can vary from yellow to brown to black27. Figure 2 shows a sample taken from swath number (1) representing partly exposed sargassum patches with most of them submerged below the water surface, as they appear in the products of the Chlor-a, SABI, 143 and 211 colour composites taken from Terra on 02 June 2006. The mean wind speed derived from QuikScat at the sample area was 4±2 ms-1. Table 1 shows the SABI, NDVI and Chlor-a values at selected points appearing in Figure 2. The SST for the bloom (not shown) did not reveal significant Proc. of SPIE Vol. 7825 782506-3 Downloaded from SPIE Digital Library on 22 Nov 2010 to 91.140.182.19. Terms of Use: http://spiedl.org/terms Figure 1. The location map of the different MODIS swaths where the SABI was tested on (left), and the TOA spectral characteristics of the surface floating species detected in relation to MODIS 500 m bands (right). temperature variability with the surrounding waters except at point C (SABI > 0), where it was flagged with poor quality. This is caused because the 3x3 pixel neighbourhood to point C were: (1) experiencing saturation or non spatial uniformity in the red band (B =678 nm) reflectance; and (2) the SST retrieval was cold relative to the reference28. At 14 point B, the SABI was negative, indicating a water column bloom; which was confirmed by the presence of a valid Chlor-a value. Clouds were distinguishable from the colour and shape that they appeared within the 143 and 211 images, although a small percentage of them were misclassified as surface algae in the SABI relation. 3.2 Nodularia spumigena– Baltic Sea Nodularia spumigena is a hepatotoxic cyanobacterium29 that can colour the water green, blue-green or yellow. Figure 3 shows a sample taken from swath number (2) of the N. spumigena bloom as it appears in the images of 211, 143, SST, SABI and Chlor-a images, taken from MODIS Aqua on 31 July 2008. The figure also shows a spectral plot of at the 500 m bands, selected points appearing in the 143 colour composite. The mean wind speed derived from QuikScat at the sample area was 5±2 ms-1. According to Figure 3, the bloom's temperature at point A was found to be 3% higher than its surrounding waters, which agrees with the findings of Kahru et al. (1993)17 and Sato et al. (1998)30, that the presence of SST 'hot spots' is an indicator of near-surface phytoplankton. Proc. of SPIE Vol. 7825 782506-4 Downloaded from SPIE Digital Library on 22 Nov 2010 to 91.140.182.19. Terms of Use: http://spiedl.org/terms Figure 2. Sargassum in the Gulf of Mexico (Terra, 02 June 2006) representing partly exposed patches and mostly submerged below the water surface as they appear in the products of the 211 and 143, Chlor-a and the SABI images in the 500 m resolution. In the Chlor-a image, point C was misclassified as cloud, and the failed Chlor-a pixels were masked in red. Figure 3. N. spumigena bloom in the Baltic Sea (Aqua, 31 July 2008) as it appears in the images of 211, 143, SST, SABI and Chlor-a . The SST at point A is 3% higher than surrounding water temperature. A spectral plot at the 500 m bands is shown for selected points appearing in the 143 colour composite. Proc. of SPIE Vol. 7825 782506-5 Downloaded from SPIE Digital Library on 22 Nov 2010 to 91.140.182.19. Terms of Use: http://spiedl.org/terms 3.3 Noctiluca miliaris– Arabian Sea Noctiluca miliaris (synonym Noctiluca scintillans Macartney) is a large cosmopolitan marine dinoflagellate31. A number of studies have tried to explain the negative relationship that exists between non-photosynthetic species like N. miliaris.a and other phytoplanktonic species by indicating the need for Noctiluca to have relatively high quantity of its phytoplankton prey to support its optimal growth32,33, during which the diatom bloom N. scintillans increases, as the algal biomass decreases34. Figure 4 shows a bloom of such species that appear in the SABI, SST, Chlor-a, 143 and the NDVI images, taken from MODIS Aqua on 18 Feb 2010. Figure 4 also shows two line transect profiles across varying aerosol layer and a sun glint region. From this plot, both the NDVI and the SABI have decreased by 15% and 3%, as the aerosol layer changed its density from low (tau_869=0.07) to high (tau_869=0.17) respectively. The sun glint transect plot shows that the SABI was less sensitive to the effect of sun glint than the NDVI. Only 2% of the failed Chlor-a pixels had a positive SABI values (floating at water surface), leaving the bulk of failed pixels having negative SABI values, that is submerged under water but close to its surface. The positive SABI pixel values were also flagged out in the SST product for the same reasons that were described earlier for the Baltic Sea area. Figure 4. N. miliaris bloom in the Arabian Sea (Aqua, 18 Feb 2010) as it appears in the, SABI, SST, Chlor-a, NDVI and 143 colour images. Two line-transect plots across varying aerosol layer (region 1) and sun glint (region 2) from (A to B) and (C to D) respectively. The failed Chlor-a and SST pixels are masked in blue and black respectively. Proc. of SPIE Vol. 7825 782506-6 Downloaded from SPIE Digital Library on 22 Nov 2010 to 91.140.182.19. Terms of Use: http://spiedl.org/terms 3.4 Ulva prolifera – Yellow Sea Ulva prolifera is unattached filamentous green macroalgae formerly known as Enteromorpha prolifera35. Figure 5 shows a bloom for the Ulva prolifera species directly under the sun glint as they appear in the SST, 211, NDVI and the SABI images on the coastline of Qingdao, in eastern China taken from MODIS Terra on 25 June 2008. Although the wind speed data was not available from QuikScat during that time, it is possible to estimate it to be not more than ~ 8 ms-1, since an increase in wind velocity above 6–8 ms-1 results in the dispersion of such aggregates17,36. The Chlor-a product could not be retrieved mainly due to sun glint, which changed the reflectance magnitude rather than the spectral shape37,38; and probably due to the high turbidity caused by the suspended sediments, which affect the distribution of planktonic and benthic organisms in the coastal waters of the Yellow Sea39,40. Figure 5. Ulva prolifera bloom in the Yellow Sea (Terra, 25 June 2008) directly under the sun glint as it appears in the SST, 211, NDVI and SABI images. 3.5 Oil spill response in the SABI and NDVI Figure 6 shows an oil spill located at (59°14' 39" N, 22°10' 29" E) in the Baltic Sea dataset, with a line transect plot across the oil spill and water (clear and turbid) shown in the NDVI and the SAB. The figure also shows a spectral plot of the corresponding oil and water at the 500 m bands. The SABI has demonstrated less sensitivity to the presence of oil and turbid water than did the NDVI. This is because spectrally a thin layer of oil (sheen) is very similar to water (optically transparent). However, this sensitivity will vary depending on the type of oil (refractive index) and texture (weathering condition). Proc. of SPIE Vol. 7825 782506-7 Downloaded from SPIE Digital Library on 22 Nov 2010 to 91.140.182.19. Terms of Use: http://spiedl.org/terms Figure 6. An oil spill located at (59°14' 39" N, 22°10' 29" E) taken from the Baltic Sea dataset. The line transect plots across an oil spill and water (clear and turbid) from (A to B) shown for the SABI and NDVI images created from TOA radiances, together with a spectral plot of the corresponding oil and water. Figure 7 shows the Gulf of Mexico oil spill taken from MODIS Aqua on 25 April 2010, where a line transect plot is shown for both the NDVI and the SABI images. In both cases, the SABI value for oil was smaller than that for surface algae. It is possible however, to discriminate between oil spills and surface algae using non-spectral methods, like for example the oil spread index (OSI) that was developed by Alawadi (2009)41. Such methods mainly rely on discriminating between oil spills and their look-alikes, like surface algae, based on the distinct texture and shape features that each make at the water surface. 3.6 Statistical analysis Figure 8 shows the SABI (sample number N=606) and NDVI (N=824) box plots for a two-class sample containing water (with or without water column bloom) and surface floating algae taken from each of the sea area datasets. The mean SABI values, appear to have higher stability in relation to the different atmospheric conditions and types of species. As in the case of the Yellow Sea, its extreme mean reflectance value could be attributed to: (1) the presence of the bloom directly under the sun glint; (2) the high Chlor-a concentration that existed in the floating Ulva prolifera species at the time; and (3) the high turbidity in the East China Sea which is a typical case II water environment, where the concentrations of phytoplankton pigments, suspended matter, and chromophoric dissolved organic matter (CDOM) are all higher than those in the open oceans42. In all of the examples, the difference between the maximum SABI values to the mean was larger than the difference from its minimum due to the abundance of the water class in the samples. The standard error estimated from the collective means of all datasets was 0.14 and 0.24 for the SABI and NDVI respectively, which again further indicates the stability of the SABI algorithm. Proc. of SPIE Vol. 7825 782506-8 Downloaded from SPIE Digital Library on 22 Nov 2010 to 91.140.182.19. Terms of Use: http://spiedl.org/terms Figure 7. An image of the Gulf of Mexico oil spill (Aqua, 25 April 2010). A line transect plot across the spill from points C to D, is shown for the NDVI and the SABI images created from TOA radiances. Figure 8. Box plots of the mean NDVI and SABI values for each of the datasets. The samples contain two classes: water (with and without bloom) and surface floating algae. 3.7 SABI AND NDVI in TOA radiance (L) t Figure 9 shows how the SABI has demonstrated better delineation results for surface floating algae than the NDVI when using TOA radiance (L) instead of reflectance. This is probably because 90% of the scattering originates from the t atmosphere particularly in the shorter visible wavelengths, and therefore, according to Equation (4), the overall SABI value will be much smaller than in the NDVI. Proc. of SPIE Vol. 7825 782506-9 Downloaded from SPIE Digital Library on 22 Nov 2010 to 91.140.182.19. Terms of Use: http://spiedl.org/terms Figure 9. The SABI (left) and NDVI (right) created from TOA radiance for the Baltic, Arabian and Yellow seas. Proc. of SPIE Vol. 7825 782506-10 Downloaded from SPIE Digital Library on 22 Nov 2010 to 91.140.182.19. Terms of Use: http://spiedl.org/terms