+44 (0)1752 874289 mhennen@argans.co.uk

Current Technologies

Earth observation of the terrestrial and marine environment is used for multiple applications, including monitoring landfills and illegal dump sites (Glanville and Chang, 2015), hazardous waste detection (Slonecker et al, 2010; Errico et al, 2014), natural disaster management (van Western, 2000), oil spill detection (Brekke and Solberg. 2005)) and marine debris tracking (Aoki et al, 2013). It is a sustainable monitoring method and provides sufficient data to gauge environmental impacts (Slonecker et al, 2010), as well as having the potential to be utilised to aid the implementation of an effective marine plastic clean-up operation. Prior research into identification of plastic on land and various other floating materials in the oceans can provide insight into potential methods for detecting plastic litter in the marine environment.

Satellite Image Pre-Processing

When using earth observation data to detect areas over land or sea various pre-processing steps are available. Synthetic Aperture Radar (SAR) images from satellites such as Sentinel-1 (S1) are radiometrically calibrated and a filter is applied to reduce speckle interference, the images are then geometrically corrected (Lu et al, 2018).

Sentinel-2 (S2) products are currently available in two main formats: Level 1C and Level 2A.  Level 1C products have radiometric and geometric corrections applied to them, including tile splitting and the calculation of top of atmosphere (TOA) reflectance values (European Space Agency, 2015). Level 2A products are further processed, for example using the Sen2Cor algorithm, to generate bottom of atmosphere (BOA) reflectance values, as well as correcting for Aerosol Optical Thickness (AOT) and Water Vapour (WV) (European Space Agency, 2015). When detecting floating debris over a body of water it is essential to use a suitable water vapour correction model (Martins et al, 2017; Pahlevan et  al, 2017).

Images taken by any Landsat (LS) satellites also require atmospheric, radiometric and geometric correction; this can be achieved with the application of modules such as  the Fast Line-of-sight Atmospheric Analysis of Spectral Hypercubes (FLAASH), or by applying coordinate systems such as the Universal Transverse Mercator (Aguilar et al, 2015; Yang et al, 2017; Kwarteng and Al-Enezi, 2011).

The growth of this UK based landfill can be seen and tracked from space (Slonecker et al, 2010).
WorldView-2 captures high resolution images that are used in conjunction with Sentinel-2 or Landsat 8 to produce panchromatic images for high resolution spectral analysis.
An example of a segmentation map depicting areas containing plastic, shown by the green-grey colour (Lu et al, 2018).

Pan-sharpening

Pan-sharpening is a process that combines a multispectral image of low spatial resolution with a panchromatic image of high spatial resolution using various fusion techniques, an example of such a technique is the application of the Kalman filter or the PANSHARP module (Garzelli and Nencini, 2007; Ehlers et al, 2010; Aguilar, 2015). Satellites such as WorldView-2 (WV2) and Quickbird-2 (QB2) can be used to provide the high spatial resolution panchromatic images with Landsat-8 (LS8) and S2 providing the lower spatial resolution multispectral images (Aguilar et al, 2015; Nemmaoui et al, 2018; Yang et al, 2017). The output of this process is a high-resolution multispectral image (Ehlers et al, 2009).

Segmentation

Segmentation models are applied to images with the aim of assigning pixels to potential categories according to the pixels spectral value (Ferro-Famil and Pottier, 2016), producing homogenous segments of specified categories (Tarantino and Figorito, 2012). For example, a study monitoring plasticulture using remote sensing could apply a model or a machine learning algorithm to separate pixels into categories such as ‘plastic’ and ‘not plastic’ (Lu et al, 2018; Nemmaoui et al, 2018; Aguilar et al, 2015). However, issues can arise as segmentation models rely on spectral values without taking spatial information into account, causing misclassification (Walter, 2004; Tarantino and Figorito, 2012). To prevent this occurring options include manually segmenting images or using a ‘supervised’ models (Aguilar et al, 2015; Walter, 2004).

Methods

Object-Based Image Analysis (OBIA) uses image segmentation and then software such as eCognition to extract the objects and classify them (Aguilar et al, 2015; Nemmaoui et al, 2018). Colour composites can be generated from satellite data sets, these can then be used to monitor and track any changes in the area of interest through photointerpretation (Kwarteng and Al-Enezi, 2011).

Some studies use spectral data extracted from a sample of pixels to build a reflectance library for all of the potential substrates within the satellite image; these are then used to classify the rest of pixels (Yang et al, 2017; Tarantino and Figorito, 2012). Satellites that have a more coarse resolution, such as LS8 and S2 contain mixed pixels so it is essential to know the exact spectral signature in all wavebands of the substrate or material of interest to establish detection criteria (Yang et al, 2017). What percentage of a mixed pixel contains the substrate/material of interest can then be determined (Yang et al, 2017). Statistically  1,732nm has been found as the best spectral feature for detecting plastic over land (Levin et al, 2007). Field work to gather in situ spectral values of various plastics allows for comparison between the recorded spectra and the surface reflectance data from the satellite image (Acuña-Ruz et al, 2018).

Creating a spectral index can also be used to identify objects in the marine and terrestrial environment (Pettorelli et al, 2005; Zha et al, 2003). More recently it has also been used to studying floating conglomerates of macroalgae in the world’s oceans for the purposes of clean-up monitoring and beaching forecasts. Several sensors have been utilised and compared to determine which is best suited for each purpose (Blondean-Passiter, 2014; Hu, 2015). Lower resolution images from satellites with a daily revisit time, e.g. the Moderate Resolution Imaging Spectroradiometer (MODIS) (250m/pixel resolution), is sufficient to obtain usable data for detecting large collections of floating vegetation (Hu et al, 2015). Higher resolution sensors are required for coastal regions as a higher detection accuracy is needed.

To determine whether a material is the material of interest the calculated index values are ran through an algorithm that compares the index value of the material to a determined threshold across several indices (Amin et al, 2009; Hu et al, 2015; Dierssen et al, 2015). The number of indices used in the algorithm is varied depending on the number of additional features and algae types that need removing from the detection criteria. A multiple index approach is then taken to verify results (Amin et al, 2009; Dierssen et al, 2015; Hu et al, 2015). Combining algorithm identification of floating material with in situ measurements can be used to determine the mass of the material present (Dierssen et al, 2015; Hu et al, 2017).

Top: Example environment being captured by Sentinel-2. Centre: Sentinel-2 has a 10m resolution, this image showcases the pixel grid overlaid over the example environment. Bottom: This showcases what happens to pixels that contain two or more surface types, the interface between the two 'mixes' within the pixel which affects an algorithms ability to classify the pixel.
The water-beach interface captured in this Sentinel-2 image highlights the mixed pixel issue.
Plastic mulched land (Lu et al, 2018).
The final, most accurate, classification map produced using the OBIA (Lu et al, 2018).

Case Studies

Martins et al (2017) used S2 images covering an area of the Amazon rainforest from S2’s multi-spectral instrument (MSI) to compare atmospheric correction algorithms. The authors (Martins et al, (2017) compared Sen2Cor, the Atmospheric Correction for OLI ‘lite’ (ACOLITE) and the Second Simulation of a Satellite Signal in the Solar Spectrum (6SV) algorithms. They (Martins et al, 2017) compared the pixel extracted reflectance values of lakes with in situ reflectance measurements by plotting the data sets and studying any trends. Of the three algorithms tested, the Sen2Cor corrected reflectance values were the closest to the in situ reflectance data. Although this study was mainly focussed on small inland lakes rather than open ocean, it demonstrates that the Sen2Cor processer used with the S2 MSI instrument is capable of producing accurate reflectance readings over water.

Lu et al (2018) investigated the use of S1 and S2 data to monitor plastic-mulched land-cover using an OBIA approach. The authors (Lu et al, 2018) used backscatter, spectral, index and texture features computed from S1 and S2 and three machine-learning classifiers (Classification and Regression Tree, Random Forest and Support Vector Machine) to detect and classify areas containing plastic-mulch. By using S1 data only they achieved an accuracy of 87.72% when detecting plastic-mulch, with the integrated S1 and S2 approach they achieved a 94.34% accuracy (Lu et al, 2018). Lu et al (2018) attributed this accuracy to the segmentation phase; after running five segmentation experiments they found that using S2’s four 10m resolution bands and the S1 modes of polarisation ‘VH_db’ and ‘VV_db’ gave the best classification results.

Nemmaoui et al (2018) used S2, LS8 and WV2 to develop an algorithm that identified crops being grown underneath plastic greenhouses. The Sen2Cor algorithm was used to gain BOA data from the S2 images and LS8 images were pan-sharpened with WV2 images (Nemmaoui et al, 2018). The authors (Nemmaoui et al, 2018) conducted field campaigns to gain an in depth understanding of the area of interest, noting things such as the number of plastic greenhouses, the crops being grown within and the total area covered by the greenhouses. They (Nemmaoui et al, 2018) used an OBIA approach to segment and pre-classify the pixels as ‘greenhouse’ (GH) and ‘non-greenhouse’ (NGH), using the mean and standard deviation of the BOA spectral data of all eight LS8 bands and all ten S2 bands.  The authors (Nemmaoui et al, 2018) found errors in the LS8 classification map when compared to the ground truth data; mixed pixels were being classed as majority GH when they were actually majority NGH. These errors occurred less frequently in the S2 classification map as S2 is a higher resolution then L8 (Nemmaoui et al, 2018).

Hu et al (2017) collected samples of Ulva prolifera and introduced 0.2kg of the algae to a tank in laboratory conditions at intervals, taking digital images after each introduction. The reflectance at each interval was recorded and used to calculate the Floating Algae Index (FAI) value (Hu et al, 2017). The authors (Hu et al, 2017) plotted this against biomass and generated a best fit curve showing that the relationship between biomass and FAI signature was linear until FAI value exceeded 0.2. Above this value the surface of the tank used in the laboratory was saturated with algae and further increases in density only generated a marginal increase in FAI (Hu et al, 2017). In situ measurements of algal reflectance followed the line of best fit calculated from the laboratory measurements (Hu et al, 2017).

A similar experiment was performed by Dierssen et al (2015) studying the abundance of Syringodium seagrass wracks off the coast of Florida. A Portable Remote Imaging Spectrometer (PRISM) was used to take aerial reflectance measurements of the Syringadium wracks. These measurements were then compared to reflectance values of Syringadium taken under laboratory conditions. Across two different days the average near infrared (NIR) reflectance was plotted against the percent cover of the wracks as observed from the air; a linear fit was observed for both. Having access to aerial imagery assisted in obtaining much more accurate measurements of the fine-scale aggregation of the wracks than would have been possible using available satellite imagery.

Yang et al (2017) used Landsat Enhanced Thematic Mapper plus images pan-sharpened with QB2 images to create a spectral library to monitor plastic greenhouses (PGH). They (Yang et al, 2017) sampled mixed pixels containing varying percentages of PGH as well as homogenous pixels containing PGH, cropland and other materials/substrates. The reflectance of a mixed pixel increased proportionally with the fraction of PGH within the pixel and an index was calculated using the blue, green, red, NIR and shortwave infrared-1, bands (Yang et al, 2017). The authors (Yang et al, 2017) found because of LS’s coarse resolution applying absolute criterion such as GH and NGH caused issues due to the number of mixed pixels. To estimate the fraction of a mixed pixel made up of PGH they developed a statistical model (Yang et al, 2017). The method and subsequent index was validated using a different and separate PGH index and by using another image data set of the same resolution (Yang et al, 2017). The authors (Yang et al, 2017) also reported that the index’s ability to perform accurately can decrease as the plastic used for the GH degrades over time.

Hu et al (2015) compared Landsat 7 (30m/pixel resolution, 16 day revisit time) to other sensors such as MODIS and the Hyperspectral Imager for the Coastal Ocean (HICO) to determine the capabilities of each in detecting different types of Sargassum. Several of these sensors image across a wide spectral range in several discrete bands, enabling calculations to be performed on the acquired reflectance values and thus generate indexes. The indexes most widely used for this purpose were Normalised Difference Vegetation Index (NDVI) and Floating Algae Index (FAI) (Hu, 2009). They take advantage of the ‘Red Edge’ present in vegetation spectra to highlight any present in a product. The authors (Hu et al, 2015) used a three-index verification system and were able to differentiate the Sargassum of interest from other algae species.

Left: Plastic greenhouses. Right: Effects of the greenhouses on the data being received by the satellite sensor (Yang et al, 2017)..
Various land cover spectra analysed by Yang et al (2017).

The monitoring of marine plastic using earth observation is in its infancy so the methods used to monitor plastic over land or other materials in the marine environment should be studied to help develop the field. The terrestrial plastic being monitored in these cases are stationary and cover vast stretches of land. It is at the opposite end of the spectrum when compared to the challenge of sensing litter in a constantly changing marine environment. However, aspects of the methodologies discussed above can be used and refined to aid marine litter detection. Carrying out field work to create suitable spectral libraries, pan-sharpening and segmenting images to classify pixels as ‘plastic’ and ‘not plastic’ can at least be attempted in the future when trying to detect plastic in the marine environment.

In all of the cases reporting on the subject using the methods above, the specific strain of algae being studied by the authors (Dierssen et al, 2015; Hu et al, 2017) was positively identified. However this was only due to the satellites they used having the correct band wavelengths for the index calculation. If an index and algorithm could be generated for plastic litter then rafts of marine plastic could potentially be identified using similar techniques on S2 products. To further that, if similar experiments to Hu et al (2015) and Deirssen et al (2015) were performed on flotsam containing plastic debris with a suitable index then more reliable estimations on the amount of plastic in identified drifts could be made.

References

Acuña-Ruz, T., Uribe,D., Taylor, R., Amézquita, L., Guzmán, M.C., Merrill, J., Martínex, P., Voisin, L. and B., C.M. (2018) Anthropogenic marine debris over beaches: Spectral characterization for remote sensing applications. Remote Sensing of Environment. Vol. 217, pp. 309-322.

Aguilar, M.A., Vallario, A., Aguilar, F.J., Lorca, A.G. and Parente, C. (2015) Object-Based Greenhouse Horticultural Crop Identification from Multi-Temporal Satellite Imagery: A Case Study in Almeria, Spain. Remote Sensing. Vol. 7(6), pp. 7378-7401.

Amin, R., Zhou, J., Gilerson, A., Gross, B., Moshary, F. and Ahmed, S. (2009) Novel optical techniques for detecting and classifying dinoflagellate Karenia brevis blooms using satellite imagery. Optical Express. Vol. 17(11), pp. 9126-9144.

Aoki, Y., Arii, M. and Koiwa, M. (2013) Spaceborne SAR data analysis for marine debris after the Great Eastern Japan Earthquake. In: ASPAR 2013: Conference Proceedings of 2013 Asia-Pacific Conference on Synthetic Aperture Radar, 23-27 Sep 2013, Tsukuba, Japan.

Blondeau-Patissier, D., Gower, J. F. R., Dekker, A. G., Phinn, S., R. and Brando, V. E. (2014) A review of ocean color remote sensing methods and statistical techniques for the detection, mapping and analysis of phytoplankton blooms in coastal  and open oceans. Process in Oceanography. Vol. 123, pp. 123-144.

Brekke C. and Solberg, A.H.S. (2005) Oil spill detection by satellite remote sensing. Remote Sensing of Environment. Vol. 95, pp. 1-13.

Dierssen, H. M., Chlus, A. and Russell, B. (2015) Hyperspectral discrimination of floating mats of seagrass wrack and the macroalgae Sargassum in coastal waters of Greater Florida Bay using airborne remote sensing. Remote Sensing of Environment. Vol. 167, pp. 247-258.

Ehlers, M., Klonus, S., Åstrand, P.J. and Rosso, P. (2010) Multi-sensor image fusion for pansharpening in remote sensing. International Journal of Image and Data Fusion. Vol. 1(1), pp. 25-45.

Errico, A. Angelino, C.V., Cicala, L., Persechino, G.. Ferrara, C., Lega, M., Vallario, A., Parente, C., Masi, G., Gaetano, R., Scarpa, G., Amitrano, D., Ruello, G., Verdoliva, L. and Poggi, G. (2015) Detection of environmental hazards through the feature-based fusion of optical and DAR data: a case study in southern Italy. International Journal of Remote Sensing. Vol. 36(13), pp. 3345-3367.

European Space Agency (2015) Sentinel-2 User Handbook. 1st ed. Retrieved from: https://sentinel.esa.int/documents/247904/685211/Sentinel-2_User_Handbook

Ferro-Famil, L. and Pottier, E. (2016) Chapter 2 – SAR imaging using Coherent Modes of Diversity: SAR Polarimetry, Interferometry and Tomography. Microwave Remote Sensing of Land Surface. pp. 67-147.

Garzelli, A. and Nencini, F. (2007) Panchromatic sharpening of remote sensing images using a multiscale Kalman filter. Pattern Recognition. Vol. 40(12), pp. 3568-3577.

Glanville, K. and Chang, H-C. (2015) Remote Sensing Analysis Techniques and Sensor Requirements to Support the Mapping of Illegal Domestic Waste Disposal Sites in Queensland, Australia. Remote Sensing. Vol. &, pp. 13053-13069.

Hu, C. (2009) A novel ocean color index to detect floating algae in the global oceans. Remote Sensing of Environment. Vol. 113, pp. 2118-2129

Hu, C., Feng, L., Hardy, R. F. and Hochberg, E. J. (2015) Spectral and spacial requirements of remote measurements of pelagic Sargassum macroalgae. Remote Sensing of Environment. Vol. 167, pp. 229-246.

Hu, L., Hu, C. and HE, M-X. (2017) Remote estimation of biomass of Ulva prolifera macroalgae in the Yellow Sea. Remote Sensing of Environment. Vol. 192, pp. 217-227.

Kwarteng, A.Y. and Al-Enezi, A. (2011) Assessment of Kuwait’s Al-Qurain Landfill Using Remotely Sensed Data. Journal of Environmental Science and Health, Part A. Vol. 39(2), pp. 351-364.

Levin, N., Lugassi, R., Ramon, O., Braun, O. and Ben-Dor, E. (2007) Remote sensing as a tool for monitoring plasticulture in agricultural landscapes. International Journal of Remote Sensing. Vol. 28(1), pp. 183-202.

Lu, L., Tao, Y. and Di, L. (2018) Object-Based Plastic-Mulched Landcover Extraction Using Integrated Sentinel-1 and Sentinel-2 Data. Remote Sensing. Vol, 10(11).

Martins, V. S., Barbosa, C. C. F., de Carvalho, L. A. S., Jorge, D. S. F., de Lucia Lobo, F. and de Moraes Novo, E. M. L. (2017) Assessment of Atmospheric Correction Methods for Sentinel-2 MSI Images Applied to Amazon Floodplain Lakes. Remote Sensing. Vol. 9, pp. 322-345.

Nemmaoui, A., Aguilar, M.A., Aguilar, F.J., Novelli, A. and Lorca, A.G. (2018) Greenhouse Crop Identification from Multi-Temporal Multi-Sensor Satellite Imagery Using Object-Based Approach: A Case Study from Almería (Spain). Remote Sensing. Vol. 10.

Pahlevan, N., Sarkar, S., Franz, B.A., Balasubramanian, S. V. and He, J. (2017) Sentinel-2 MultiSpectral Instrument (MSI) data processing for aquatic science applications: Demonstrations and validations. Remote Sensing of Environment. Vol. 201, pp. 47-56.

Pettorelli, N., Olav Vik, j., Mysterud, A., Gaillard, J., Tucker, C. J. and Stenseth, N. C. (2005) Using the satellite-derived NDVI to assess ecological responses to environmental change. Trends in Ecology and Evolution. Vol. 20(9), pp. 503-510.

Slonecker, T. Fisher, G.B., Aiello, D.P. and Haack, B. (2010) Visible and Infrared Remote Imaging of Hazardous Waste: A Review. Remote Sensing. Vol. 2, pp. 2474-2508.

Tarantino, E. and Figorito. B. (2012) Mapping Rural Areas with Widespread Plastic Covered Vineyards Using True Colour Aerial Data. Remote Sensing. Vol. 4, pp. 1913-1928.

van Western, C.J. (2000) Remote Sensing for Natural Disaster Management. International Archives of Photogrammetry and Remote Sensing. Vol. 33(B7), pp. 1609-1617.

Walter, V. (2004) Object-based classification of remote sensing data for change detection. ISPRS Journal of Photogrammetry and Remote Sensing. Vol. 58(3-4), pp. 225-238.

Yang, D., Chen, J., Zhou, Y., Chen, X., Chen, X. and Cao, X. (2017) Mapping plastic greenhouse with medium spatial resolution satellite data: Development of a new spectral index. ISPRS Journal of Photogrammetry and Remote Sensing. Vol. 128, pp. 47-60.

Zha, Y., Gao, J. and Ni, S. (2003) Use of normalized difference built-up index in automatically mapping urban areas from TM imagery. International Journal of Remote Sensing. Vol. 24(3), pp. 583-594.