Ele Vahtmäe

Mapping coloured dissolved organic matter concentration in coastal waters

Optical properties of the Baltic Sea are dominated by coloured dissolved organic matter (CDOM). High concentration of CDOM is probably one of the reasons why standard chlorophyll-retrieval algorithms fail badly in the Baltic Sea. Our aim was to test can CDOM be mapped by remote sensing instruments in coastal waters of relatively CDOM-rich environments like the Baltic Sea. The results show that sensors with high radiometric resolution, such as Advanced Land Imager (ALI), can be used for mapping CDOM in a wide concentration range. The ALI image also showed that optical properties of coastal waters are extremely variable. CDOM concentration may vary 4–5-fold in two adjacent 30 m pixels. This indicates a need for relatively high spatial resolution (30 m or less) remote sensing data in monitoring coastal environments.

Classifying the Baltic Sea Shallow Water Habitats Using Image-Based and Spectral Library Methods

The structure of benthic macrophyte habitats is known to indicate the quality of coastal water. Thus, a large-scale analysis of the spatial patterns of coastal marine habitats enables us to adequately estimate the status of valuable coastal marine habitats, provide better evidence for environmental changes and describe processes that are behind the changes. Knowing the spatial distribution of benthic habitats is also important from the coastal management point of view. A big challenge in remote sensing mapping of benthic habitats is to define appropriate mapping classes that are also meaningful from the ecological point of view. In this study, the benthic habitat classification scheme was defined for the study areas in the relatively turbid north-eastern Baltic Sea coastal environment. Two different classification methods—image-based and the spectral library—method were used for image classification. The image-based classification method can provide benthic habitat maps from coastal areas, but requires extensive field studies. An alternative approach in image classification is to use measured and/or modelled spectral libraries. This method does not require fieldwork at the time of image collection if preliminary information about the potential benthic habitats and their spectral properties, as well as variability in optical water properties exists from earlier studies. A spectral library was generated through radiative transfer model HydroLight computations using measured reflectance spectra from representative benthic substrates and water quality measurements. Our previous results have shown that benthic habitat mapping should be done at high spatial resolution, owing to the small-scale heterogeneity of such habitats in the Estonian

Predicting Species Cover of Marine Macrophyte and Invertebrate Species Combining Hyperspectral Remote Sensing, Machine Learning and Regression Techniques

In order to understand biotic patterns and their changes in nature there is an obvious need for high-quality seamless measurements of such patterns. If remote sensing methods have been applied with reasonable success in terrestrial environment, their use in aquatic ecosystems still remained challenging. In the present study we combined hyperspectral remote sensing and boosted regression tree modelling (BTR), an ensemble method for statistical techniques and machine learning, in order to test their applicability in predicting macrophyte and invertebrate species cover in the optically complex seawater of the Baltic Sea. The BRT technique combined with remote sensing and traditional spatial modelling succeeded in identifying, constructing and testing functionality of abiotic environmental predictors on the coverage of benthic macrophyte and invertebrate species. Our models easily predicted a large quantity of macrophyte and invertebrate species cover and recaptured multitude of interactions between environment and biota indicating a strong potential of the method in the modelling of aquatic species in the large variety of ecosystems.

Relating remotely sensed optical variability to marine benthic biodiversity.

Biodiversity is important in maintaining ecosystem viability, and the availability of adequate biodiversity data is a prerequisite for the sustainable management of natural resources. As such, there is a clear need to map biodiversity at high spatial resolutions across large areas. Airborne and spaceborne optical remote sensing is a potential tool to provide such biodiversity data. The spectral variation hypothesis (SVH) predicts a positive correlation between spectral variability (SV) of a remotely sensed image and biodiversity. The SVH has only been tested on a few terrestrial plant communities. Our study is the first attempt to apply the SVH in the marine environment using hyperspectral imagery recorded by Compact Airborne Spectrographic Imager (CASI). All coverage-based diversity measures of benthic macrophytes and invertebrates showed low but statistically significant positive correlations with SV whereas the relationship between biomass-based diversity measures and SV were weak or lacking. The observed relationships did not vary with spatial scale. SV had the highest independent effect among predictor variables in the statistical models of coverage-derived total benthic species richness and Shannon index. Thus, the relevance of SVH in marine benthic habitats was proved and this forms a prerequisite for the future use of SV in benthic biodiversity assessments.

Detecting patterns and changes in a complex benthic environment of the Baltic Sea

Today, the knowledge on the distribution of marine habitats is very fragmented and temporal changes in such patterns are even less known. In this study we assessed spatial variability and temporal dynamics of benthic habitat types in a relatively turbid northeastern Baltic Sea coastal environment using the space-borne multispectral sensor QuickBird. Seven broad habitat classes were defined for the study area representing the most typical habitats of the coastal environment. The studied classes were bare sand, the brown alga Fucus vesiculosus, hard bottom with ephemeral algae, higher-order plants and/or charophytes on soft bright bottom, dense higher-order plant habitats, and drifting algal mats and deep water (>3 m). Two QuickBird images acquired over a 3 year interval (2005 to 2008) of Western-Estonian archipelago were processed and change detection analysis applied. Although there was a relatively large scatter in reflectance variability within each habitat type, the analyses allowed a clear differentiation of most habitat types. Exceptions were the lack of statistical differences among deep water, drifting algae, and dense higher-order plant communities, as well as among low density higher-order plant and algal communities. Major changes in the spatial patterns of benthic habitats occurred in hydrodynamically active areas. Differences in water properties caused some confusion in classification and therefore resulted in inaccuracies in maps of change. Thus, the used broad habitat classes represent the limit of the method and the multispectral sensors do not allow finer elements of habitats to be captured.

Mapping Baltic Sea shallow water environments with airborne remote sensing

It is known that the structure of benthic macrophyte and invertebrate habitats indicate the quality of coastal water. Thus, a large-scale analysis of the spatial patterns of coastal marine habitats makes it possible to adequately estimate the status of valuable coastal marine habitats, provide better evidence for environmental changes, and describe the processes behind the changes. Knowing the spatial distribution of benthic habitats is also important from the coastal management point of view. Our previous results clearly demonstrated that remote sensing methods can be used to map water depth and distribution of taxonomic groups of benthic algae (e.g., red, green, and brown algae) in the optically complex coastal waters of the Baltic Sea. We have as well shown that benthic habitat mapping should be done at high spatial resolution owing to the small-scale heterogeneity of such habitats in Estonian coastal waters. Here we tested the capability of high spatial resolution hyperspectral airborne image in its application for mapping benthic habitats.A big challenge is to define appropriate mapping classes that are also meaningful from the ecological point of view. In this study two benthic habitat classification schemes—broader level and finer level—were defined for the study area. The broader level classes were relatively well classified, but discrimination among the units of the finer classification scheme posed a considerable challenge and required a careful approach. Benthic habitat classification provided the highest accuracy in the case of the Spectral Angle Mapper classification method applied to a radiometrically corrected image. Further processing levels, such as spatial filtering and glint correction, decreased the classification accuracy.

Airborne mapping of shallow water bathymetry in the optically complex waters of the Baltic Sea

Abstract. Accurate determination of the water depth is important for marine spatial planning, producing maritime charts for navigation, seabed morphology studies, and carrying out different activities in the coastal waters. Bathymetric data are lacking foremost in the shallow water regions as those areas are often inaccessible to the hydrographic ships carrying out echo sounding measurements. Remote sensing technology can be used as an alternative for shallow water bathymetry mapping. Varieties of empirical methods have been proposed for bathymetry retrieval, where the relationship between remotely sensed radiance of the water body and the water depth at sampled locations was established empirically. Two most widely used depth derivation methods, the linear band model proposed by Lyzenga (1978, 1985, 2006), and the log-transformed band ratio model proposed by Stumpf et al. (2003), were applied to the different preprocessing level airborne Hyspex hyperspectral images from the optically complex Baltic Sea area and evaluated for accuracy. Results showed that the Lyzenga linear band model outperformed the Stumpf log-transformed band ratio model. The best results were achieved with the atmospherically corrected images. The application of glint correction did not improve, but even reduced the accuracy of bathymetric maps.

Predicting macroalgal pigments (chlorophyll a, chlorophyll b, chlorophyll a + b, carotenoids) in various environmental conditions using high-resolution hyperspectral spectroradiometers

ABSTRACT Photosynthetic pigments may indicate the health and productivity of vegetation and thereby are among the most important targets of the remote-sensing science. We studied the relationship between macroalgae pigment concentration measured in situ and spectral reflectance, to develop predictive remote-sensing methods for macroalgal pigments. The measurements of spectral reflectance of macroalgae were made using both a field portable spectrometer Ramses built by TriOS GmbH (Germany) and a laboratory hyperspectral imaging device HySpex built by Norsk Elektro Optikk (Norway). Our results showed that differences in total chlorophyll (Chl-a + b) concentrations resulted in the consistent change of spectral reflectance for studied brown (Fucus vesiculosus) and green (Cladophora glomerata, Ulva intestinalis) macroalgae species. Charophytes (Chara aspera, Chara horrida) were also studied, and the relationship was much weaker for this taxon. If spectral indices predicted relatively well the concentration of Chl-a + b (R2 = 0.64–0.73) and the carotenoid to total chlorophyll ratio (Car:Chl-a + b, R2 = 0.80) across the five studied macroalgae species, then the concentration of chlorophyll a (Chl-a), chlorophyll b (Chl-b), and carotenoids (Car) were more difficult to model (R2 = 0.004–0.51). The HySpex imaging system yielded systematically better results in predicting pigment concentrations compared to the Ramses spectroradiometer. By using traditional assessment of pigment concentration along with the Hyspex imaging device, we were able to build models with a capability to predict the spatial patterns of pigment concentration for Baltic Sea macroalgae.

Sun glint correction of airborne AISA images for mapping shallow-water benthos

Airborne sensors have higher spatial and spectral resolution than satellite sensors, providing greater accuracy in benthic habitat mapping in case of high spatial heterogeneity. Nevertheless, the effect of wave-induced sun glint may obscure the radiance originating from within the water. Such glint is particularly noticeable due to the high spatial resolution of the sensor and may impede mapping of benthic features. This paper describes the application of the sun glint correction schemes on to airborne hyperspectral AISA measurements acquired on the area of the West-Estonian archipelago during the campaign in July 2006. Currently proposed sun glint removal procedures assume zero water leaving signal in near infrared part of spectrum. This assumption is not true in waters less than about 2 m deep where part of the water leaving signal is originated from the bottom. As a result the shallow water pixels are overcorrected during glint removal procedure and the shapes of reflectance spectra are distorted. This has serious implications on shallow water bottom classification results, especially if spectral libraries of in situ measured or modelled reflectance spectra are used in classification of remote sensing imagery. Therefore, it is important to preserve spectral signatures of these areas if sun glint removal is necessary. We propose an alternative sun glint removal procedure where the amount of glint in each pixel is estimated from the depth of oxygen absorption feature at 760 nm relative to a baseline. The new method removes sun glint successfully and at the same time preserves the shape and magnitude of shallow water reflectance spectra.

Mapping Seagrass Biomass with Photo-Library Method

Validation of benthic habitat maps produced from remote sensing imagery is quite time consuming and expensive. Validating maps of seagrass biomass is even more sophisticated and time consuming as the seagrass has to be collected, dried and weighed in the laboratory. We developed a method for estimating the dry weight of the seagrass based on photo transect data and a photo library of quadrats with known seagrass biomass. For seagrass biomass estimation we selected 13 different bottom classes. A photo of each 25 × 25 cm quadrat was taken prior to collecting the samples for each class. Seagrass (and macroalgae, if present) dry weight for each class was measured in the laboratory. These photos were then used to estimate seagrass biomass in 100 m long geolocated photo transects. Seagrass dry weight estimated from the photo transects using the photo-library method was compared with QuickBird satellite radiances. Preliminary results show that QuickBird imagery may be used for mapping seagrass biomass even in highly variable environment such as the Ngederrak Reef in Palau.