Analysis of classification accuracy for pre-filtered multichannel remote sensing data

Noise is one of the main factors degrading the quality of original multichannel remote sensing data and its presence influences classification efficiency, object detection, etc. Thus, pre-filtering is often used to remove noise and improve the solving of final tasks of multichannel remote sensing. Recent studies indicate that a classical model of additive noise is not adequate enough for images formed by modern multichannel sensors operating in visible and infrared bands. However, this fact is often ignored by researchers designing noise removal methods and algorithms. Because of this, we focus on the classification of multichannel remote sensing images in the case of signal-dependent noise present in component images. Three approaches to filtering of multichannel images for the considered noise model are analysed, all based on discrete cosine transform in blocks. The study is carried out not only in terms of conventional efficiency metrics used in filtering (MSE) but also in terms of multichannel data classification accuracy (probability of correct classification, confusion matrix). The proposed classification system combines the pre-processing stage where a DCT-based filter processes the blocks of the multichannel remote sensing image and the classification stage. Two modern classifiers are employed, radial basis function neural network and support vector machines. Simulations are carried out for three-channel image of Landsat TM sensor. Different cases of learning are considered: using noise-free samples of the test multichannel image, the noisy multichannel image and the pre-filtered one. It is shown that the use of the pre-filtered image for training produces better classification in comparison to the case of learning for the noisy image. It is demonstrated that the best results for both groups of quantitative criteria are provided if a proposed 3D discrete cosine transform filter equipped by variance stabilizing transform is applied. The classification results obtained for data pre-filtered in different ways are in agreement for both considered classifiers. Comparison of classifier performance is carried out as well. The radial basis neural network classifier is less sensitive to noise in original images, but after pre-filtering the performance of both classifiers is approximately the same.

An intercomparison of bio-optical techniques for detecting dominant phytoplankton size class from satellite remote sensing

Satellite remote sensing of ocean colour is the only method currently available for synoptically measuring wide-area properties of ocean ecosystems, such as phytoplankton chlorophyll biomass. Recently, a variety of bio-optical and ecological methods have been established that use satellite data to identify and differentiate between either phytoplankton functional types (PFTs) or phytoplankton size classes (PSCs). In this study, several of these techniques were evaluated against in situ observations to determine their ability to detect dominant phytoplankton size classes (micro-, nano- and picoplankton). The techniques are applied to a 10-year ocean-colour data series from the SeaWiFS satellite sensor and compared with in situ data (6504 samples) from a variety of locations in the global ocean. Results show that spectral-response, ecological and abundance-based approaches can all perform with similar accuracy. Detection of microplankton and picoplankton were generally better than detection of nanoplankton. Abundance-based approaches were shown to provide better spatial retrieval of PSCs. Individual model performance varied according to PSC, input satellite data sources and in situ validation data types. Uncertainty in the comparison procedure and data sources was considered. Improved availability of in situ observations would aid ongoing research in this field.

Macroscale factors affecting diatom abundance: a synergistic use of Continuous Plankton Recorder and satellite remote sensing data

Diatoms exist in almost every aquatic regime; they are responsible for 20% of global carbon fixation and 25% of global primary production, and are regarded as a key food for copepods, which are subsequently consumed by larger predators such as fish and marine mammals. A decreasing abundance and a vulnerability to climatic change in the North Atlantic Ocean have been reported in the literature. In the present work, a data matrix composed of concurrent satellite remote sensing and Continuous Plankton Recorder (CPR) in situ measurements was collated for the same spatial and temporal coverage in the Northeast Atlantic. Artificial neural networks (ANNs) were applied to recognize and learn the complex non-monotonic and non-linear relationships between diatom abundance and spatiotemporal environmental factors. Because of their ability to mimic non-linear systems, ANNs proved far more effective in modelling the diatom distribution in the marine ecosystem. The results of this study reveal that diatoms have a regular seasonal cycle, with their abundance most strongly influenced by sea surface temperature (SST) and light intensity. The models indicate that extreme positive SSTs decrease diatom abundances regardless of other climatic conditions. These results provide information on the ecology of diatoms that may advance our understanding of the potential response of diatoms to climatic change.

Enhancement of primary production in the North Atlantic outside of the spring bloom, identified by remote sensing of ocean colour and temperature

The heterogeneity in phytoplankton production in the North Atlantic after the spring bloom is poorly understood. We analysed merged microwave and infrared satellite sea surface temperature (SST) data and ocean colour phytoplankton size class biomass, primary production (PP) and new production (ExP) derived from SeaWiFS data, to assess the spatial and temporal frequency of surface thermal fronts and areas of enhanced PP and ExP. Strong and persistent surface thermal fronts occurred at the Reykjanes Ridge (RR) and sub-polar front (SPF), which sustain high PP and ExP and, outside of the spring bloom, account for 9% and 15% of the total production in the North Atlantic. When normalised by area, PP at the SPF is four times higher than the RR. Analysis of 13 years of satellite ocean colour data from SeaWiFS, and compared with MODIS-Aqua and MERIS, showed that there was no increase in Chla from 1998 to 2002, which then decreased in all areas from 2002 to 2007 and was most pronounced in the RR. These time series also illustrated that the SPF exhibited the highest PP and the lowest variation in Chla over the ocean colour record. This implies that the SPF provides a high and consistent supply of carbon to the benthos irrespective of fluctuations in the North Atlantic Oscillation.

Remote sensing of assimilation number for marine phytoplankton

Estimating primary production at large spatial scales is key to our understanding of the global carbon cycle. Algorithms to estimate primary production are well established and have been used in many studies with success. One of the key parameters in these algorithms is the chlorophyll-normalised production rate under light saturation (referred to as the light saturation parameter or the assimilation number). It is known to depend on temperature, light history and nutrient conditions, but assigning a magnitude to it at particular space-time points is difficult. In this paper, we explore two models to estimate the assimilation number at the global scale from remotely-sensed data that combine methods to estimate the carbon-to-chlorophyll ratio and the maximum growth rate of phytoplankton. The inputs to the algorithms are the surface concentration of chlorophyll, seasurface temperature, photosynthetically-active radiation af the surface of the sea, sea surface nutrient concentration and mixed-layer depth. A large database of in situ estimates of the assimilation number is used to develop the models and provide elements of validation. The comparisons with in situ observations are promising and global maps of assimilation number are produced.