Ahmed El-Habashi

Satellite Retrievals of Karenia brevis Harmful Algal Blooms in the West Florida Shelf Using Neural Networks and Comparisons with Other Techniques

We describe the application of a Neural Network (NN) previously developed by us, to the detection and tracking, of Karenia brevis Harmful Algal Blooms (KB HABs) that plague the coasts of the West Florida Shelf (WFS) using Visible Infrared Imaging Radiometer Suite (VIIRS) satellite observations. Previous approaches for the detection of KB HABs in the WFS primarily used observations from the Moderate Resolution Imaging Spectroradiometer Aqua (MODIS-A) satellite. They depended on the remote sensing reflectance signal at the 678 nm chlorophyll fluorescence band (Rrs678) needed for both the normalized fluorescence height (nFLH) and Red Band Difference algorithms (RBD) currently used. VIIRS which has replaced MODIS-A, unfortunately does not have a 678 nm fluorescence channel so we customized the NN approach to retrieve phytoplankton absorption at 443 nm (aph443) using only Rrs measurements from existing VIIRS channels at 486, 551 and 671 nm. The aph443 values in these retrieved VIIRS images, can in turn be correlated to chlorophyll-a concentrations [Chla] and KB cell counts. To retrieve KB values, the VIIRS NN retrieved aph443 images are filtered by applying limiting constraints, defined by (i) low backscatter at Rrs 551 nm and (ii) a minimum aph443 value known to be associated with KB HABs in the WFS. The resulting filtered residual images, are then used to delineate and quantify the existing KB HABs. Comparisons with KB HABs satellite retrievals obtained using other techniques, including nFLH, as well as with in situ measurements reported over a four year period, confirm the viability of the NN technique, when combined with the filtering constraints devised, for effective detection of KB HABs.

Remote estimation of in water constituents in coastal waters using neural networks

Remote estimations of oceanic constituents from optical reflectance spectra in coastal waters are challenging because of the complexity of the water composition as well as difficulties in estimation of water leaving radiance in several bands possibly due to inadequacy of current atmospheric correction schemes. This work focuses on development of a multiband inversion algorithm that combines remote sensing reflectance measurements at several wavelengths in the blue, green and red for retrievals of the absorption coefficients of phytoplankton, color dissolved organic matter and nonalgal particulates at 443nm as well as the particulate backscatter coefficient at 443nm. The algorithm was developed, using neural networks (NN), and was designed to use as input measurements on ocean color bands matching those of the Visible Infrared Imaging Radiometer Suite (VIIRS). The NN is trained on a simulated data set generated through a biooptical model for a broad range of typical coastal water parameters. The NN was evaluated using several statistical indicators, initially on the simulated data-set, as well as on field data from the NASA bio-Optical Marine Algorithm Data set, NOMAD, and data from our own field campaigns in the Chesapeake Bay which represent well the range of water optical properties as well as chlorophyll concentrations in coastal regions. The algorithm was also finally applied on a satellite - in situ databases that were assembled for the Chesapeake Bay region using MODIS and VIIRS satellite data. These databases were created using in-situ chlorophyll concentrations routinely measured in different locations throughout Chesapeake Bay and satellite reflectance overpass data that coexist in time with these in-situ measurements. NN application on this data-sets suggests that the blue (412 and 443nm) satellite bands are erroneous. The NN which was assessed for retrievals from VIIRS using only the 486, 551 and 671 bands showed that retrievals that omitted the 671 nm band was the most effective, possibly indicating an inaccuracy in the VIIRS 671 band that needs to be further investigated.

Satellite retrievals of Karenia brevis harmful algal blooms in the West Florida shelf using neural networks and impacts of temporal variabilities

Abstract. We apply a neural network (NN) technique to detect/track Karenia brevis harmful algal blooms (KB HABs) plaguing West Florida shelf (WFS) coasts from Visible-Infrared Imaging Radiometer Suite (VIIRS) satellite observations. Previously KB HABs detection primarily relied on the Moderate Resolution Imaging Spectroradiometer Aqua (MODIS-A) satellite, depending on its remote sensing reflectance signal at the 678-nm chlorophyll fluorescence band (Rrs678) needed for normalized fluorescence height and related red band difference retrieval algorithms. VIIRS, MODIS-A’s successor, does not have a 678-nm channel. Instead, our NN uses Rrs at 486-, 551-, and 671-nm VIIRS channels to retrieve phytoplankton absorption at 443 nm (aph443). The retrieved aph443 images are next filtered by applying limits, defined by (i) low Rrs551-nm backscatter and (ii) a minimum aph443 value associated with KB HABs. The filtered residual images are then converted to show chlorophyll-a concentrations [Chla] and KB cell counts. VIIRS retrievals using our NN and five other retrieval algorithms were compared and evaluated against numerous in situ measurements made over the four-year 2012 to 2016 period, for which VIIRS data are available. These comparisons confirm the viability and higher retrieval accuracies of the NN technique, when combined with the filtering constraints, for effective detection of KB HABs. Analysis of these results as well as sequential satellite observations and recent field measurements underline the importance of short-term temporal variabilities on retrieval accuracies.

Characterization of sun and sky glint from wind ruffled sea surfaces for improved estimation of polarized remote sensing reflectance

During two cruises in 2014, the polarized radiance of the ocean and the sky were continuously acquired using a HyperSAS-POL system. The system consists of seven hyperspectral radiometric sensors, three of which (one unpolarized and two polarized) look at the water and similarly three at the sky. The system autonomously tracks the Sun position and the heading of the research vessel to which it is attached in order to maintain a fixed relative azimuth angle with respect to the Sun (i.e. 90°) and therefore avoid the specular reflection of the sunlight. For the duration of both cruises, (NASA Ship Aircraft Bio-Optical Research (SABOR), and NOAA VIIRS Validation/Calibration), in situ inherent optical properties (IOPs) were continuously acquired using a set of instrument packages modified for underway measurement, and hyperspectral radiometric measurements were taken manually at all stations. During SABOR, an underwater polarimeter was deployed when conditions permitted. All measurements were combined in an effort to first develop a glint (sky + Sun) correction scheme for the upwelling polarized signal from a wind driven ocean surface and compare with one assuming that the ocean surface is flat.

Impact of fluorescence on the underwater polarized light field: comparison of theory and field measurements

We have examined, in earlier work, the relationship between naturally induced chlorophyll-a fluorescence and the underwater polarized oceanic light field. This shows the un-polarized fluorescence causes a reduction in the degree of polarization over the fluorescence spectral range. Theory shows that the peak of the reduction in polarization occurs at or near the fluorescence peak. Furthermore, it also shows that the magnitude of this reduction in degree of polarization can be related to both the magnitude of the fluorescence as well as the intensity of the underwater light field over the fluorescence spectral range. To examine this relationship in detail, a vector radiative transfer code (VRTE) for the coupled atmosphere-ocean system was employed for a variety of oligotrophic and eutrophic water conditions. The VRTE used measured inherent optical properties (IOPs) for these water conditions as inputs to simulate the complete elastic and inelastic components of the underwater light field, as well as the degree of linear polarization (DoLP) associated with it. These theoretical predictions were then compared with the results of DoLP measurements carried out using by our multiangular hyperspectral polarimeter. A comparison of the measured reduction in degree polarization of the underwater light field over the fluorescence spectral range, and the magnitude of the fluorescence causing it, confirmed the validity of our theoretical relationship, and the feasibility of determining the natural fluorescence existing in an underwater light field from polarization measurements.

Multi-band algorithms for the estimation of chlorophyll concentration in the Chesapeake Bay

Standard blue-green ratio algorithms do not usually work well in turbid productive waters because of the contamination of the blue and green bands by CDOM absorption and scattering by non-algal particles. One of the alternative approaches is based on the two- or three band ratio algorithms in the red/NIR part of the spectrum, which require 665, 708, 753 nm bands (or similar) and which work well in various waters all over the world. The critical 708 nm band for these algorithms is not available on MODIS and VIIRS sensors, which limits applications of this approach. We report on another approach where a combination of the 745nm band with blue-green-red bands was the basis for the new algorithms. A multi-band algorithm which includes ratios Rrs(488)/Rrs(551)and Rrs(671)/Rrs(745) and two band algorithm based on Rrs671/Rrs745 ratio were developed with the main focus on the Chesapeake Bay (USA) waters. These algorithms were tested on the specially developed synthetic datasets, well representing the main relationships between water parameters in the Bay taken from the NASA NOMAD database and available literature, on the field data collected by our group during a 2013 campaign in the Bay, as well as NASA SeaBASS data from the other group and on matchups between satellite imagery and water parameters measured by the Chesapeake Bay program. Our results demonstrate that the coefficient of determination can be as high as R2 > 0.90 for the new algorithms in comparison with R2 = 0.6 for the standard OC3V algorithm on the same field dataset. Substantial improvement was also achieved by applying a similar approach (inclusion of Rrs(667)/Rrs(753) ratio) for MODIS matchups. Results for VIIRS are not yet conclusive.

Analysis of polarimetric image by full stokes vector imaging camera for retrieval of target polarization in underwater environment

Polarized image of underwater light field contains rich information of and the targets strongly affected by the water inherent optical properties. We present a comprehensive analysis of the polarimetric images of a manmade underwater target with known polarization properties acquired by a full Stokes vector imaging camera in underwater environment. The effects of the camera’s parameters such as numerical aperture and orientation are evaluated. With the knowledge acquired in the analysis of such a forward polarimetric imaging process, a method for retrieval of the inherent optical properties of the water and the target polarization is explored.

Impact on satellite retrievals of temporal changes in Karenia brevis harmful algal blooms in the West Florida Shelf

We examine the impact of temporal changes on satellite retrievals of Karenia brevis Harmful Algal blooms (KB HABS) in the West Florida Shelf (WFS). These impacts are compared for retrievals from both VIIRS and MODIS-A using a number of retrieval techniques. The comparisons include our recently developed neural network (NN) technique. The neural network, previously developed by us, was trained on 10,000 data point part of a synthetic data of 20,000 inherent optical properties (IOPs) based on a wide range of IOP parameters for a large variety of natural conditions based on the NOMAD data. The NN then uses as inputs the remote sensing reflectance (Rrs) measurements at 486, 551 and 671 and 488, 555 and 667 nm which are available from VIIRS and MODIS-A respectively, to retrieve phytoplankton absorption at 443 nm in satellite images. This information, when combined with backscatter information has been shown by us to be effective for obtaining retrievals of KB HB HABs in the WFS. Other retrieval algorithms included in the present comparison are the blue/green OCI/OC3, the Generalized Inherent Optical Property (GIOP) model and the Quasi-Analytical Algorithm, (QAA version 5). The accuracies of VIIRS retrievals using all five techniques were then compared against all the in-situ measurements available over the 2012-2016 period for which concurrent or near concurrent match ups could be obtained with VIIRS retrievals. Analysis of retrieval statistics showed that the NN technique achieved the best accuracies. The analysis highlighted the impact of temporal variabilities on retrieval accuracies. The results showed the importance of having a shorter overlap time window between in-situ measurement and satellite retrieval. Retrievals where the maximum permissible overlap time window was shortened to 15 minutes, exhibited very significantly improved retrieval accuracies over those that were obtained with a 100 minute overlap time window, Retrievals that relaxed the overlap time window between in-situ measurement and satellite retrievals to simply the same day were hopelessly inaccurate by comparison. These results are believed to reflect the impact of temporal variabilities on retrieval accuracies. They underline the time limitations associated with satellite retrievals of inherently variable conditions and a changing scene. The temporal effects associated with KB HABs retrievals in the WFS were also examined by using images from consecutive VIIRS - MODIS-A – VIIRS overpasses overlapping the same parts of the WFS containing KB blooms. The consecutive images, all within an approximately 100 minute period, appear to confirm the changing bloom features. The temporal behavior of the KB blooms in the WFS over short time periods (less than100 minuntes) was further confirmed by a recent set of in-situ field measurements off the coast of Sarasota.

Ocean surface characterization using snapshot hyperspectral polarimetric imager

Results of measurements by a novel snapshot hyperspectral polarimetric imager are presented using several data sets acquired from ocean platforms. Based on the unique availability of the pixel-to-pixel total, sky and water leaving radiances at multiple wavelengths, variations of these parameters for wind-roughened surface are assessed and possible errors in measurements of these parameters are estimated. Measurements made by the imager are compared with coincident ones from the green-band SALSA Stokes vector imaging camera, a push-broom hyperspectral polarimetric imager operated by Naval Research Laboratory (NRL), and with simulations using a vector radiative transfer code, all demonstrating excellent agreement.

Satellite monitoring and interpretation of dynamically changing concentrations of Karenia Brevis harmful algal blooms in the West Florida Shelf (Conference Presentation)

Karenia brevis Harmful Algal blooms (KB HABS) plague the coasts of the West Florida Shelf (WFS) and effective monitoring is needed to provide information to local authorities for health warnings etc. We compare results of satellite retrievals of KB HABs, using previously existing algorithms, for both MODIS and VIIRS, as well as with our more recently developed neural network (NN) algorithm. Retrievals are compared against in-situ HABs measurements. To obtain sufficient numbers of in-situ measurements nearly concurrent with satellite overpasses, comparisons were extended over 2012-17. Algorithms compared included nFLH, RBD OCI/OC3, GIOP and QAA. Retrieval statistics showed that the NN technique achieved the best accuracies, possibly due to the fact that it uses the 486, 551 and 671 nm channels for the retrievals, which are less impacted by atmospheric correction inadequacies in coastal waters, than the deeper blue channels used with other retrieval algorithms. Results highlight impacts of temporal variabilities on retrieval accuracies. Thus a shorter overlap time window between in-situ measurement and satellite observation of 15 minutes, showed significantly better accuracies than a 100 minutes, reflecting short-term changes in the KB HABs scene being observed. These relatively rapid temporal changes are further confirmed by retrievals from consecutive satellite overpasses: VIIRS-MODIS-A –VIIRS (second overpass) within a 100 minute period, and by in-situ measurements in the WFS. Temporal changes are seen to clearly affect the timeliness and relevance of satellite retrievals of HABs and Ocean Color parameters, particularly in coastal zones with dynamically changing conditions, and need to be taken into account, including possible development of alternate observation means in real time, such as UAVs.