Satellite detection of harmful algal blooms (HABs)

Mati Kahru,


“Standard” ocean color products such as chlorophyll-a concentration (Chl-a, e.g. O’Reilly et al., 1998) are not specific to HABs but are useful for detecting very high Chl-a levels that are often associated with HABs. For example, the very high Chl-a in the Santa Barbara channel (SBC) in Figures 1 and 2 may be or may not be not associated with toxic blooms. The bloom in Fig. 2 was most likely dominated by a toxic diatom Pseudo-nitzschia. Similar blooms in SBC were described in Anderson et al (in press) and were associated with significant marine mammal deaths in 2003.

Fig. 1. MODIS-Aqua Chl-a image of January 29, 2004 showing an upwelling bloom off Point Conception. The small red circles on the image are the standard stations of a CalCOFI cruise. For a full-resolution version click on the image or here. This image has been created using Chl-a with a color look-up table over sea and quasi-true color over land and clouds. Fig. 2. MODIS-Aqua Chl-a image of April 18, 2007 of the Southern California Bight.

 Going beyond Chl-a and detecting individual species, toxic or nontoxic, or phytoplankton groups has long been a goal of many researchers. Most of the existing algorithms of detecting phytoplankton types are empirical and limited to certain bio-optical conditions or geographic domains. For example, near-surface accumulations the cyanobacterium Nodularia spumigena in the Baltic Sea can be reliably detected by their high reflectance (Kahru et al., 1994), open ocean blooms of cyanobacteria Trichodesmium (Subramaniam et al., 2002) and coccolithophores (Brown et al., 1994) can be detected by the modified ocean reflectance spectra. Lower specific absorption of diatoms can be used to detect diatom-dominated blooms (Sathyendranath et al., 2004). Dinoflagellate blooms with high concentrations of the mycosporine-like amino acids (MAAs) can be detected by their reduced reflectance in the UV bands (Kahru and Mitchell, 1998). High-concentration cyanobacteria blooms can be detected by the presence of phycocyanin (Simis et al, 2005; Vincent et al., 2004).

However, none of the methods listed above can be used to detect HABs or even those same phytoplankton types with currently available satellite data in the coastal zone where they are easily fooled by a combination of factors including suspended sediments. One can easily confirm that the Trichodesmium and Coccolithophore algorithms are not reliable in the coastal zone by looking at the Level-2 flags of almost any standard ocean color Level-2 image in turbid waters: the corresponding flags are often set but in most cases there is no in situ evidence of the Trichodesmium or Coccolithophore blooms ever being detected in these areas.

A more fundamental approach for detecting phytoplankton groups is to invert the spectrum of water leaving radiance using a library of spectral signatures of individual species or groups of species (Roesler et al., 2004). Unfortunately the current space-borne ocean color sensors do not have the spectral resolution that would make this approach even theoretically possible (Dierssen et al., 2006). It is well known that the red coloration of the sea surface during red tides is not specific to absorption properties of any particular group of phytoplankton and is caused by a combination of a number of factors, e.g. the strong backscattering of a dense aggregation of cells, the reduction in the relative importance of water absorption due to the shallow surface distribution of the bloom and even the physiology of the human visual system (Roesler et al., 2004, Dierssen et al., 2006). It is also possible to try to invert the currently available and spectrally low-resolution ocean color data into inherent optical properties (IOPs) and that can be used to infer the existence of different phytoplankton groups. This approach is closely related to the semi-analytic algorithms of Chl-a and other water constituents. The well-known algorithms in this category are the GSM (Maritorena et al., 2001) and the QAA (Lee et al., 2002, 2007) which have been incorporated into the SeaDAS software and are therefore easy to apply. While these algorithms are theoretically superior to the standard maximum band ratio algorithm (OC3m, O’Reilly et al., 1998), in practice they are very sensitive to the errors in atmospheric correction of individual bands. For example, instead of the smooth and quite realistic distribution of Chl-a with OC3m (Fig. 2) we often get distorted images with GSM and Carder et al. (1999) algorithms. This is happening because of errors in the standard atmospheric correction cause bands 412 and 443 to be unrealistically low and even negative in the SBC. The maximum band ratio algorithm recovered from these errors by switching to longer wavelength bands but the semianalytic algorithms are rendered practically useless in complex near shore conditions. Even without the problems in atmospheric correction the semianalytic GSM algorithm is not well suited for inverting the radiances in these optically complex waters (Kostadinov et al., 2007).

Fig. 3. Surface accumulations of cyanobacteria Nodularia in the Baltic Sea as detected by enhanced "true-color" MODIS imagery on July 30, 2003. For the full-resolution click here.


While the 1 km resolution looks great compared to a typical oceanographic cruise, it is usually not sufficient to assess the distribution of a typical harmful algal bloom in a coastal bay. Some HABs cover large areas and can be detected with standard ocean color imagery. For example, the annual midsummer blooms of a toxic cyanobacterium Nodularia spumigena in the Baltic Sea produce large-scale accumulations of biomass at the surface and are easily detectable with almost any visible imagery due to their strong reflectance. Time series of these accumulations have been constructed using images from various satellites (Kahru et al., 1994; Kahru, 1997; Kahru et al., 2000). The spectacular quasi-true color image of MODIS-Aqua from July 30, 2003 in Fig. 3 shows the Nodularia surface accumulation in the middle of the Baltic Sea between the Swedish island of Gotland on the left and the Estonian Island of Saaremaa on the right. The small island visible near the top is Gotska Sandö. These surface accumulations act like tracers of surface currents and show the spectacular eddy fields present in the sea. The distance between the large islands (the easternmost point of Gotland and the westernmost point of Saaremaa) is about 150 km (81 miles). The scale of these accumulations extends into hundreds of kilometers. At the resolution of 250 m the image shows incredible details but the extent of the accumulations can be mapped using relatively low-resolution imagery. While the cyanobacterium Nodularia is toxic and can cause the death of various animals these accumulations mostly occur in the open sea and their direct economic effect on human activities is relatively small. They have an important role in the ecosystem by assimilating gaseous nitrogen (nitrogen fixation) and adding nitrogen into the ecosystem.


While standard ocean color products can be used to map large-scale HABs, typical HABs occur in coastal environments at much smaller scales. First, the typical pixel size of 1 km much too large for detecting coastal HABs. Second, the difficulties in atmospheric correction and radiance inversion in coastal areas severely limit the usefulness of standard ocean color products for HAB detection.  In a typical case the pixels represented by a coastal HAB are just blocked out because of a failure at some point of the standard processing algorithms. MODIS has medium-resolution (250 and 500 m) bands but these bands were not designed for ocean applications and lack standard ocean processing routines. However, with careful processing these bands can provide valuable ocean applications (Hu et al., 2003, 2004).


A harmful phytoplankton bloom dominated by a dinoflagellate Gymnodinium sanguineum in Paracas Bay, Peru in April, 2004 caused estimated economic damage of $28.5 million. While standard ocean color products of SeaWiFS and MODIS were of little use in this case due to insufficient resolution and problems in atmospheric correction and radiance inversion, MODIS medium-resolution bands provided valuable information with empirical processing algorithms (Kahru et al., 2004).

Fig. 4 from Kahru et al., 2004 shows that the application of two empirical products in monitoring of the devastating bloom in Paracas Bay. The left column shows that the true-color (red-green-blue) images using, respectively, MODIS bands 1, 4, and 3 can clearly identify the distribution of the bloom in the Bay by its conspicuous bright color. The right column shows the turbidity index, a semi-quantitative measure of the amount of particulate material in the near-surface water. Darker areas show higher turbidity. Julian day is shown for the true-color images, and the corresponding date (month/day/year) is shown for the turbidity images. While turbidity is not specific to algal blooms, it is a quantitative estimate of the intensity of the bloom once the existence of the bloom is detected by the true-color images. During the rise and fall of the bloom in the bay turbidity was inversely correlated with oxygen content in the bay. Oxygen depletion caused most of the damage to the benthic communities. The top panel (A) shows the bloom in the increasing phase, panel B shows the maximum extent of the bloom, panel D shows the decreasing phase of the bloom and the bottom panel (E) shows the normal conditions after the bloom.


 Fig. 4. Enhanced "true-color" and turbidity images of a devastating HAB in Paracas Bay, Peru (from Kahru et al., 2004). For the full-resolution version click here.



A compromise approach is to use the MODIS medium resolution bands at the top of the atmosphere, apply a simplified atmospheric correction (that does not result in negative water-leaving radiances) and create enhanced true-color images. This method cannot unequivocally detect HABs, for known surface HABs it can produce reliable space and time distributions. The 250 m turbidity product has been shown to correlate well with the biomass of a HAB and inversely with the oxygen concentration in the affected bay (Kahru et al., 2004). While this “medium resolution” method of Kahru et al. (2004) can produce only semi-quantitative turbidity estimates and rather subjective estimates of the main characteristics of water masses, compared to the other methods available for routine application to satellite data it is a good compromise. As shown in Fig. 5, it separates the high backscatter waters (green) that are often associated with high suspended sediments or HABs from the highly absorbing waters (dark, almost black) that are due to high concentrations of phytoplankton pigments in the water column. According to water samples off Scripps Pier analyzed by Melissa Carter the bloom (“red tide”) was made up mostly of Prorocentrum micans and a few other dinoflagellates, Ceratium divaricatium, Ceratium fusus, and Gymnodinium sanguineum (syn. Akashiwo sanguinea).  The measured chlorophyll concentration off Scripps Pier on 4/19/2007 was 14.96 mg/m3. The processed moderate resolution method using showed that the bloom extended along the coast of Southern California and beyond. However, the bloom area was relatively narrow, about only 1.5-3 km from the coast while the band of increased Chl-a was much wider. The eddy-like feature off San Clemente (2nd red arrow from top) extended 19 km from the coast and seemed to transport the dinoflagellate bloom offshore. Ocean color data from SeaWiFS and MODIS is available at 1 km resolution and that is too coarse for mapping the booms in the coastal zone. Semianalytic algorithms (e.g. GSM01, QAAv4) have the potential to invert radiances into inherent optical properties such as absorption and scattering at different wavelengths but typically fail in the coastal zone due to problems in the atmospheric correction. Much more work is needed until reliable inversion of the reflectances in the coastal zone becomes routinely possible. In the meantime, using the MODIS L1B data gives a possibility of semi-quantitative mapping of blooms and other optically active constituents in the water at 250 m resolution. Both enhanced “true color” and turbidity products are useful for mapping HABs. This method was successfully applied to monitor a devastating HAB in Paracas Bay in Peru (Kahru et al., 2004). Fig. 2A is an enhanced “true color” image of MODIS Aqua that shows the extent of the dinoflagellate bloom (green areas along the coast, red arrows). Areas of high Chl-a appear dark in this image due to strong absorption by Chl-a (white arrows). Standard Chl-a images from both SeaWiFS and Aqua (standard band ratio method) with 1 km resolution show that the dark areas correspond to high Chl-a. The coefficient of backscattering at 440 nm determined with the QAA method is also shown. Inversion methods are very sensitive to errors in the water leaving radiances and therefore the coastal zone has been masked.


Fig. 5. The same MODIS Aqua scene as in Fig. 2. 

(A) Enhanced “true color” image processed with the moderate resolution algorithm (Kahru et al., 2004). For a full-resolution version click on the image or here.

(B) Turbidity using the MODIS 250 m bands (Kahru et al., 2004). For a full-resolution version click on the image or here.


The processed satellite images can be conveniently distributed over the web and visualized and navigated using Google Earth. The sample “true color” image in Google Earth format is available from the following link

While some knowledge of the HABs from in situ sampling is necessary, this empirical method can be the best current option to extend the in situ observations in space and time. Considering the fact that MODIS L1B data is available daily from two sensors (Terra from 1999 and Aqua from 2002) makes this very powerful and underused method in the study of HABs and other coastal phenomena. Up to 2 passes per day (both MODIS Aqua and Terra) are available. That is a great advantage compared to high resolution/narrow swath sensors such as ASTER and Landsat TM which can cover a local area only once in 2 weeks or even less due to clouds.





Anderson, C.A.., D.A. Siegel, M.A. Brzezinski, N. Guillocheau, Controls on Temporal Patterns in Phytoplankton Community Structure in the Santa Barbara Channel, California, J Geophys Res, Oceans, 2007, in press.

Brown, C.W. and J.A. Yoder. Coccolithophorid blooms in the global ocean. J. Geophys. Res., 99, 7467-7482.

Carder, K. L., Chen, F. R., Lee, Z. P., Hawes, S., & Kamykowski, D., Semi-analytic MODIS algorithms for chlorophyll a and absorption with bio-optical domains based on nitrate-depletion temperatures, J Geophys Res, 104(C3), 5403-5421, 1999.

Dierssen, H.M., R.M. Kudela, J.P. Ryan and R.C. Zimmerman, Red and black tides: Quantitative analysis of water-leaving radiance and perceived color for phytoplankton, colored dissolved organic matter, and suspended sediments, Limnol. Oceanogr., 51(6), 2646-2659.

Hu, C., F. Muller-Karger, C. J. Taylor, D. Myhre, B. Murch, A.L. Odriozola, and G. Godoy (2003), MODIS detects oil spills in Lake Maracaibo, Venezuela, Eos Trans. AGU, 84(33), 313.

Hu, C., Z. Chen, T.D. Clayton, P. Swarnzenski, J.C. Brock, and F.E. Muller-Karger, Assessment of estuarine water-quality indicators using MODIS medium-resolution bands: Initial results from Tampa Bay, FL. Remote Sens. Environ., 93: 423-441, 2004.

Kahru, M., U. Horstmann, O. Rud, Satellite detection of increased cyanobacteria blooms in the Baltic Sea: Natural fluctuation or ecosystem change? Ambio, 23 (8):  469-472, 1994.

Kahru, M., B.G. Mitchell.  Spectral reflectance and absorption of a massive red tide off Southern California,  J. Geophys. Res. Vol. 103, No. C10, 21,601-21,609, 1998.

Kahru, M., B.G. Mitchell, A. Diaz, M. Miura. MODIS Detects a Devastating Algal Bloom in Paracas Bay, Peru. EOS, Trans. AGU, Vol. 85, N 45, p. 465-472, 2004.

Kahru, M., B.G. Mitchell, A. Diaz, Using MODIS medium-resolution bands to monitor harmful algal blooms, Proc. of SPIE, Vol. 5885 (SPIE, Bellingham, WA, 2005) • 0277-786X/05/$15 • doi: 10.1117/12.615625, 2005, 6 p.

Kahru, M.  Using satellites to monitor large-scale environmental change:  A case study of cyanobacteria blooms in the Baltic Sea.  In:  Monitoring algal blooms:  New techniques for detecting large-scale environmental change.  M. Kahru and Ch. W. Brown (Eds.).  43-61, 1997.

Kostadinov, T.S., D.A. Siegel, S. Maritorena, N. Guillocheau, Ocean color observations and modeling for an optically complex site: Santa Barbara Channel, California, USA,  J Geophys Res, Oceans, 112, C07011, 10.1029/2006JC003526, 2007.

Lee, Z. P., K. L. Carder, and R. Arnone, Deriving inherent optical properties from water color: A multi-band quasi-analytical algorithm for optically deep waters, Appl. Opt., 41, 5755–5772, 2002.

Lee, ZhongPing, A. Weidemann, J. Kindle, R. Arnone, K. L. Carder, and C. Davis, Euphotic zone depth: Its derivation and implication to ocean-color remote sensing, J Geophys Res, 112, C03009, doi:10.1029/2006JC003802, 2007.

Maritorena, S., D. A. Siegel, and A. R. Peterson, Optimization of a semianalytical ocean color model for global-scale applications, Appl. Opt., 41, 2705– 2714, 2002.

O'Reilly, J.E., S. Maritorena, B.G. Mitchell, D.A. Siegel, K.L. Carder, S.A. Garver, M. Kahru and C.R. McClain.  Ocean color chlorophyll algorithms for SeaWiFS.  J. Geophys. Res. Vol. 103, No.C11, p. 24,937-24,953, 1998.

Roesler, C. S., S. M. Etheridge and G. C. Pitcher. 2004. Application of an ocean color algal taxa detection model to red tides in the Southern Benguela, pp.303-305. In: Steidinger, K. A., Lansdberg, J. H., Tomas, C.R., and Vargo, G. A. [eds.]. Harmful Algae 2002. Florida Fish and WildlifeConservation Commission, Florida Institute of Oceanography, andIntergovernmental Oceanographic Commission of UNESCO.

Sathyendranath, S., L. Watts, E.Devred, T. Platt, C. Caverhill, and H. Maass. Disrcimination of diatoms from other phytoplankton using ocean color data. Mar. Ecol. Prog. Ser., 272, 59-68, 2004.

Simis, S.G.H, S.W.M. Peters, and H.J. Gons, Remote sensing of the cyanobacterial pigment phycocyanin in turbid inland water, Limnol. Oceanogr., 50(1), 237–245, 2005.

Subramaniam, A., C.W. Brown, R.R. Hood, E.J. Carpenter, and D.G. Capone. Detecting Trichodesmium blooms in SeaWiFS imagery, Deep-Sea Res. II, 49, 107-121, 2002.

Vincent, R. K., X. M. Qin, R. M. L. McKay, J. Miner, K. Czajkowski, J. Savino, and T. Bridgeman, Phycocyanin detection from LANDSATTM data for mapping cyanobacterial blooms in Lake Erie, Rem. Sens. Environ. 89: 381–392, 2004.