A sun glint correction method for hyperspectral imagery containing areas with non-negligible water leaving NIR signal

https://doi.org/10.1016/j.rse.2009.06.016Get rights and content

Abstract

Present sun glint removal methods overcorrect data collected in very shallow (less than 2 m) waters if the sensors used do not have bands in far infrared part of the spectrum. The reason is assuming of zero water leaving signal at near infrared (NIR) wavelengths. This assumption is not valid in very shallow waters, but also in areas where aquatic vegetation reaches water surface and in case of phytoplankton blooms that reach very high biomass or form surface scum. We propose an alternative method that can be used for successful glint removal even if the sensor does not have spectral bands beyond 800 nm. The proposed method utilises the presence and depth of the oxygen absorption feature near 760 nm as an indicator of glint contamination. This method allows removing sun glint from hyperspectral imagery preserving shape and magnitude of reflectance spectra in the cases where the negligible water leaving NIR signal is not valid.

Introduction

Hyperspectral airborne sensors provide information with high spectral and spatial resolution. Such information can provide great benefits for mapping submerged benthic vegetation, coral reef habitats, seagrass beds etc. in coastal areas where the spatial heterogeneity is high. Use of airborne imagers is often the most cost effective method to map water quality in lakes and geomorphologically complex coastal waters where satellite sensors cannot provide adequate spatial and/or spectral resolution. However, airborne remote sensing can be seriously impeded by the effect of wave-induced sun glint. When water surface is not flat, the direct radiance originating from the sun can be reflected on the crests or slopes of waves. The reflected radiance does not contain any information about the water constituents and benthic features. Sun glint effect is often a factor in wide-field-of-view acquisition airborne or satellite missions, despite acquisition time and solar/viewing geometry optimised to avoid glint (Hochberg et al., 2003). Contribution from glint may be reduced by having the sensor flown towards or away from the sun, but it is not always possible due to the characteristics of a field site or scheduling (Mustard et al., 2001). Therefore the ability to recognise and remove contributions from sun glint is required.

Sun glint shows strong spatial variations that require individual calculation and correction for each pixel depending on the water state (Cavalli et al., 2006). Glint correction models have been derived for open ocean waters by Cox and Munk, 1954, Cox and Munk, 1956. Their approach is reasonable for satellite sensors with a wide spatial resolution, but may not be well suited in case of coastal waters and high spatial resolution (Heege & Fischer, 2000). Heege and Fischer (2000) proposed a sun glint removal procedure that uses radiative transfer model inversion and a SWIR (short-wavelength infrared) band beyond 1200 nm (usually 1600–1800 nm) where the water absorption is very high and water leaving signal is negligible. However, high spatial resolution satellite sensors (IKONOS, QuickBird) and many airborne imagers (CASI, AISA) do not have any spectral bands beyond 900 nm or 1000 nm.

Hochberg et al. (2003) proposed a method for removal of sea surface effects from high-resolution imagery in shallow coastal environments. The method is based on two assumptions: (1) as near infrared (NIR) light is strongly absorbed by water, any signals remaining in near infrared channels (after atmospheric correction) represent light reflected from the sea surface, and (2) the amount of sun glint in the NIR band is linearly related to the glint contribution in the visible bands.

While performing the sun glint correction on airborne remotely sensed data of coastal environment, some limitations have to be taken into account (Cavalli et al., 2006). Referred sun glint removal procedure assumes zero water leaving signal in near infrared part of the spectrum. In practice, however, there is always some “residual” radiance in the NIR part of the spectrum (Hochberg et al., 2003). The NIR signal can be significant if the water is optically shallow. For example benthic algal cover (Schweizer et al., 2005, Vahtmäe et al., 2006), seagrasses (Dekker et al., 2005, Mumby et al., 1997, Pasqualini et al., 2005) and corals (Hochberg et al., 2004, Holden and LeDrew, 2002, Kutser et al., 2003, Kutser and Jupp, 2006) have high reflectance in NIR part of the spectrum. Model estimates show that the water leaving signal in NIR part of the spectrum is significant if the water depth is less than 2 m (Kutser et al., 2003, Vahtmäe et al., 2006). Plants have high reflectance in NIR and SWIR parts of the spectrum. Therefore, the water reflectance is high in these parts of the spectrum if aquatic vegetation reaches water surface and is covered only by thin film of water. In this case the reflectance values may become negative and the shapes of reflectance spectra are distorted if the sun glint removal procedures assume zero water leaving signal in NIR. Coral reef tops are often closer to the water surface than 2 m and lagoons are often shallower. Near shore zone with waters less than 2 m deep is quite extensive in many waterbodies (e.g. the Baltic Sea and many lakes). Kelp forests and seagrass beds are potential sources of overcorrection as plants often reach water surface (especially in low tide). Some phytoplankton groups (e.g. cyanobacteria) form dense subsurface blooms and surface scum. In the latter cases the reflectance spectrum is similar to that of terrestrial vegetation (Jupp et al., 1994, Kutser, 2004) i.e. high in NIR and SWIR. Therefore, the areas where the zero in NIR (and SWIR) signal assumption does not work in sun glint removal may be extensive and usually these parts of images are of the greatest interest.

Hedley et al. (2005) have shown that the method by Hochberg et al. (2003) is sensitive to outlier pixels and requires time consuming masking of land and clouds. They proposed a revised method where the glint intensity is obtained based on the large amount of pixels instead of two. Establishing the linear relationship between a NIR band and each visible band allows the removal of the glint contribution.

It has been shown that modelled (or measured) spectral libraries (look-up tables) can be used to map shallow water bottom types and water depth (Kutser et al., 2006a, Kutser and Jupp, 2002, Lesser and Mobley, 2007, Louchard et al., 2003, Mobley et al., 2005) or phytoplankton biomass in bloom conditions (Kutser, 2004). If the spectral libraries are compared with image spectra using procedures like Spectral Angle Mapper then the absolute reflectance values are not critical. It is only important to preserve the right shape of reflectance spectra while pre-processing tasks, like glint removal, are carried out. Uniqueness of the obtained results may be a problem if only the shape of reflectance spectra is taken into account. For example reflectance spectra of algal blooms and seagrass beds are similar in their shape. Lesser and Mobley (2007) have shown that uniqueness of the solution is not a problem when both the shape and magnitude of reflectance spectra are taken into account. However, it must be noted here that taking into account both the spectral shape and magnitude of reflectance spectra does not guarantee uniqueness of the solution. This is especially true in case of sensors with limited spectral resolution (Kutser et al., 2006b) or when water depths and bottom types in the study site are extremely variable.

Physics based image interpretation methods require high quality reflectance input whether they use the spectral library approach (Kutser, 2004, Kutser et al., 2006a, Kutser and Jupp, 2002, Lesser and Mobley, 2007, Louchard et al., 2003, Mobley et al., 2005) or model inversion techniques (Brando et al., 2009). Currently used glint removal procedures (Hedley et al., 2005, Hochberg et al., 2003) cause overcorrection of glint in some circumstances and consequently reduce the image area that can effectively be used. In the present paper we propose an alternative sun glint removal procedure for hyperspectral imagery that preserves reflectance spectra from overcorrection if the NIR part contains detectable amount of water leaving signal (shallow water, plants reaching water surface, phytoplankton blooms) and SWIR data is not available.

Section snippets

Image data

Airborne measurements were carried out in the West Estonian Archipelago (Baltic Sea) using AISA (Airborne Imaging Spectrometer for Applications) spectrometer made by Specim Ltd. in Finland. The flight was carried out on July 26, 2006. The day was quite windy. Wind speeds were between 7 and 12 m/s in closest meteostations. As a result sun glint was a problem in several areas of the image. Many unsheltered areas were also covered with white caps and we observed also stripes of foam on the water

Results

Visual inspection of the glint removal products (Fig. 1) shows that all three methods improve the image. The new method we propose produces slightly noisier results. Some reasons of the relative noisiness are described in the Discussion part of the manuscript. However, it is more important to test how well the different methods perform in preserving spectral information after the glint removal procedure. It is known that the method by Hochberg et al. (2003) overcorrects shallow water imagery in

Discussion

Comparison of the three glint removal procedures shows that the new method is performing better than the other two in case of very shallow (less than 2 m) water and when aquatic plants are reaching water surface. There are plenty of situations where the use of the new method is preferable in sun glint removal from hyperspectral imagery. Those include coastal and reef areas with waters less than 2 m deep, areas where kelp forests, seagrass or macroalgae reach water surface or where aquatic

Conclusions

We have developed a new sun glint removal method for hyperspectral data that allows to correct imagery preserving the shape and magnitude of reflectance spectra in particularly challenging cases where the SWIR data is not available and other glint removal procedures fail. These cases include:

  • very shallow (less than 2 m) waters,

  • areas where aquatic vegetation reaches water surface,

  • areas with lot of exposed sea bottom (mudflats, seagrass beds, etc.),

  • areas with very strong subsurface phytoplankton

Acknowledgements

The study was supported by Estonian Science Foundation grant No. 6051, Estonian Environmental Monitoring Programme and Estonian basic research grant 0712699s05, and Estonian National Monitoring Program. We would like to thank the anonymous reviewers whose comments helped to improve the manuscript from its original version.

References (32)

  • CavalliR.M. et al.

    Correction on sun glint effect on MIVIS data of the Sicily campaign in July 2000

    Annals of Geophysics

    (2006)
  • CoxC. et al.

    Measurements of the roughness of the sea surface from photographs of the sun's glitter

    Journal of Optical Society of America

    (1954)
  • CoxC. et al.

    Slopes of the sea surface deduced from photographs of the sun's glitter

    Bulletin of the Scripps Institution of Oceanography

    (1956)
  • HedleyJ.D. et al.

    Simple and robust removal of sun glint for mapping shallow-water benthos

    International Journal of Remote Sensing

    (2005)
  • HeegeT. et al.

    Sun glitter correction in remote sensing imaging spectrometry

  • HochbergE.J. et al.

    Sea surface correction of high spatial resolution IKONOS images to improve bottom mapping in near-shore environments

    IEEE Transactions on Geoscience and Remote Sensing

    (2003)
  • Cited by (86)

    View all citing articles on Scopus
    View full text