1. Introduction
Phytoplankton are micro-autotrophs that play a major role in primary production and oxygen generation in aquatic systems. However, a disproportional increase in phytoplankton biomass may result in algal blooms. There are certain species of phytoplankton that produce bio-toxins (e.g.,
Karenia brevis) [
1,
2]. Proliferation of these species, also called harmful algal blooms (HAB), causes serious impact on marine and human health [
1]. Nutrient inputs from contributing watersheds and rivers have been shown to be responsible for the accelerated eutrophication of receiving water bodies [
3,
4]. The problem of algal bloom is expected to worsen with global climate change [
5]. Understanding the dynamics of phytoplankton populations and their distribution enables assessment of the nutrient state, health, and ecological integrity of a body of water. Since chlorophyll-a (Chl-a) exists in every species of phytoplankton [
6], its concentration is used as a proxy for the distribution of phytoplankton biomass [
7,
8]. The conventional method of Chl-a estimation requires water sample collection and laboratory analysis [
9]. This method, effort-intensive and time consuming, is unsuitable for large spatio-temporal scales. Instead, satellite-based sensors can be used for the synoptic assessment of Chl-a at large spatial scales. The launch of the first satellite borne ocean color sensor, the Coastal Zone Color Scanner (CZCS), in 1978, began an evolution of satellite deployments with improved sensors, higher precision, and an increased number of spectral bands [
10]. Currently, one operational ocean color sensor, Moderate Resolution Imaging Spectroradiometer (MODIS) Aqua, collects data with 1–2 days of temporal resolution. The default Chl-a retrieving algorithm for MODIS Aqua, the Ocean Color 3M (OC3M) algorithm, is a blue-green band ratio algorithm [
11].
In spite of the development of advanced and precise sensors, the error in the satellite estimation of Chl-a concentration in coastal waters is high [
12]. This can be due to the bottom reflectance in shallow water systems. Researchers have classified the ocean water areas as case 1 and case 2 water. The optical property of the surface of deep ocean water is dominated by phytoplankton and is termed as case 1 water [
13]. In coastal regions, the optical property of water is influenced by colored dissolved organic matter (CDOM), bottom reflectance, and total suspended matter (TSM), and is referred to as case 2 water. The blue-green band ratio strongly correlates to Chl-a concentration in case 1 water, however, in case 2 waters, the correlation becomes weak [
14]. Furthermore, because of a low attenuating tendency, the green band is heavily influenced by bottom reflectance in shallow coastal water [
15]. The OC3M algorithm that uses the blue-green band ratio has been shown to yield accurate results in case 1 waters [
16]. However, the algorithm overestimates the Chl-a in case 2 waters [
12].
Over 50% of the world’s population lives in coastal zones [
17], and coastal waters provide many ecosystem services of human importance including fisheries and recreation. Primary production in coastal areas influences fisheries, eutrophication, and algal blooms that affect human populations. Ocean color data from a satellite-based sensor are the only practical tools for the global assessment of spatiotemporal variation in phytoplankton populations. The long record of MODIS ocean color data of coastal regions currently cannot be utilized due to a lack of a precise algorithm for the Chl-a estimation. A robust algorithm would make use of all available data and will have a significant effect on the understanding of various factors that regulate primary production in ocean water. Furthermore, precise assessment of phytoplankton biomass will assist in understanding models of atmospheric carbon dioxide flux to the ocean, and the influence of anthropogenic contaminants on the marine ecosystem [
9].
The red band has a tendency of attenuating within surface water depths, and therefore is less affected by bottom reflectance [
15]. Furthermore, it is less sensitive to CDOM [
18]. Therefore, red bands have been used in several studies to develop Chl-a algorithm for shallow coastal water [
15,
16,
19,
20,
21]. In such algorithms, the green band or near-infrared (NIR) band are commonly used in combination with red band [
15,
16,
19,
20,
21]. Gons, et al. [
22] used an algorithm based on backscattering coefficients at NIR bands to retrieve the Chl-a concentration. Gilerson, et al. [
18] used an algorithm based on the ratio of a red to NIR band. Gitelson, et al. [
23] and Le, et al. [
24] demonstrated the applicability of red-NIR algorithms in estuarine water, including the Chesapeake Bay. Blakey, et al. [
25] developed the Benthic Class Specific algorithm to reduce the noise due to bottom reflectance. However, these algorithms have a limited application. Processed data with atmospheric correction are not provided for NIR bands as part of the standard data suite. Applicability of the Benthic Class Specific Algorithm is contingent on the availability of Sea Grass Density data at the location. To utilize the abundance of MODIS ocean color data of coastal regions, an improved algorithm is required that will use the wave bands for which MODIS reflectance data are available.
MODIS Aqua’s level-2 data suit provides processed reflectance data in blue, green and red bands. Therefore, many algorithms have been developed for MODIS sensor using green and red bands, and have been shown to perform better than the OC3M algorithm but significant inaccuracy in estimation using these algorithms leaves the scope for further improvement [
19,
20,
21]. A green-red band ratio algorithm is based on the fact that Chl-a reflects radiation in green band whereas absorbs radiations in the red band resulting in a reflectance peak in the green region and reflectance trough in the red region of the spectrum [
14]. Schalles, et al. [
14] in his mesocosm experiment demonstrated the wavelength position of the green peak shifts towards the higher wavelength as the Chl-a concentration increases. Similarly, the trough observed in red band shifts from lower band to higher band as the Chl-a concentration increases. Therefore, instead of using the ratio of reflectance at one wavelength, each from green and red bands, an algorithm that employs the ratio of maximum reflectance in green band to minimum reflectance in red band could further improve the satellite estimation of Chl-a in the coastal water (case 2) by more efficiently targeting the peak and trough, and increasing the reflectance ratio.
In this study, the performance of the OC3M algorithm in case 1 and case 2 waters was assessed, and an algorithm was developed for case 2 water, using an alternative band combination from the green-red wavebands of the MODIS Aqua sensor. Since the algorithm utilizes four bands from the green and red region of the spectrum, it has been named as Green-Red Ocean Color 4 algorithm (here after GROC4). The performance of the GROC4 algorithm was tested and compared with that of the OC3M and other green-red algorithms using an independent dataset, and finally the seasonal performance of the GROC4 was evaluated. Thus, the specific objectives of this study were to (1) evaluate the performance of the OC3M model in case 1 and case 2 waters, (2) develop an improved Chl-a estimation algorithm for case 2 water, and (3) evaluate the performance of the newly developed model in case 2 water with seasonal data.
5. Conclusions
In this study, the performance of the OC3M algorithm was assessed across the deep ocean and a coastal water body. An improved algorithm was developed based on reflectance data from the green and red bands and its performance in the coastal water system of Chesapeake Bay was tested using an independent dataset. The OC3M algorithm worked well for case 1 water. However, the error of estimation was very high in case 2 waters. The result of this analysis demonstrates that OC3M algorithm is useful for synoptic mapping of Chl a in the deep ocean region. However, the high error of estimation in Chesapeake Bay shows that the algorithm is unsuitable for satellite estimation of Chl-a in coastal waters.
The present study also demonstrated the usefulness of maximum-green-red band ratio formulation for Chl a estimation in complex coastal water. The GROC4 algorithm performed significantly better than the OC3M algorithm in the coastal water of Chesapeake Bay. The RMSE was reduced from of 24.783 mg m−3 to 4.924 mg m−3 when the GROC4 algorithm was used instead of the OC3M algorithm for the same validation dataset. The evaluation of seasonal performance of the algorithm demonstrated that the GROC4 algorithm is significantly correlated with Chl-a during spring and summer.
The implementation of the algorithm developed in this study is simple and requires only reflectance data in wavebands that are available from MODIS Aqua sensor. It could be used to understand the dynamics of Chl-a in coastal water by using the long record of publicly available MODIS Aqua ocean color data. Although, GROC4 algorithm reduced the error in deriving Chl-a by a significant margin, the following limitations remain. The matchup dataset used for the development of the algorithm is small (134 pairs). For most of the pixels in the MODIS imagery of the study area, reflectance data is not available, and frequency of in situ observations is low. It would be better to utilize a larger set of matchup pixels in order to develop an algorithm with a greater accuracy. Furthermore, the satellite estimation of Chl-a in coastal water is severely affected by the operational atmospheric correction procedures that are known to be inefficient [
46]. Therefore, an encouraging prospect for enhancement in satellite estimation of Chl-a in coastal water is improvement in atmospheric correction procedure and development of an algorithm using the maximum green-red band ratio with a larger set of matchup pixels.
This study addresses the important issue of coastal water algal bloom mapping [
47] using satellite data with reasonable accuracy, and showed an improvement over the existing operational algorithm. The results of this study can be applicable to water bodies and can lead to further improvements in as algorithm performance.