Nov 7, 2014 - One topic is general change detection, where real landscape transitions are determined by measuring .... optimization problem: given a multi-dimensional input signal s tÐ ÑâªÒs1 tÐ Ñ; â¦; sn tÐ Ñâ¼T, ..... matrix is obtained b
Mealiness in apples can be assessed by sensory panels with sensorial ..... that only less than 2% apples stored in cold storage had developed mealiness, while ...
Nov 20, 2015 - Detection of Lettuce Discoloration Using Hyperspectral. Reflectance Imaging. Changyeun Mo 1, Giyoung Kim 1, Jongguk Lim 1, Moon S. Kim 2, ...
Hyperspectral Imaging. Abstract â The focus of this project was to develop and implement detection algorithms for imag- ing and non-imaging spectroscopy ...
May 11, 2011 - Keywords: Meteorological Database, Data Mining, Clustering, Self Organizing .... interfaces, and software libraries that help read and write scientific data files. .... Component Planes (CPs) can be drawn for each input feature.
It is important that robots be able to identify the state of the environment in which ..... of automatically selecting vision thresholds by automatically identifying the ...
HYPERSPECTRAL ENDMEMBER DETECTION USING MORPHOLOGICAL. AUTOASSOCIATIVE MEMORIES. By. DANIEL S. MYERS. A THESIS PRESENTED ...
 L.Bruzzone and S.B.Serpico, âAn iterative technique for the detection of land-cover transitions in multitemporal remote- sensing images,â IEEE Trans.Geosci.
Mar 17, 2010 - use covariance shrinkage or the dominant mode rejection approximation. .... estimation of Î´, is thoroughly discussed by Ledoit and Wolf.7. 3.
Nov 18, 2011 - including temperature, dissolved oxygen, and electrical conductivity. This was a complex analysis ...... Yokogawa Tech Rep. Engl Ed. p. 13-. 17.
a novel approach for detection and discrimination of gaseous plumes in IR hyperspectral ..... M. Hinnrichs and B. Piatek, âHand held hyperspectral imager.
for semisupervised hyperspectral image classification, by introducing .... like in convolution network architecture is given in Eq. (2). 1. 1. 1 ... the next path layer.
Feb 7, 2002 - (X = 0.565fj,m) for (a) Class 2 Grass and (b) Mixed back- ground data. Also shown are straight lines corresponding to the normal distributions. 17. IX ..... Neyman-Pearson detectors are specified by the same LR function; they only diffe
A major drawback of most of the existing hyperspectral anomaly detection methods ... complexity with the state-of-the-art representation learning method, namely ...
Feb 11, 2016 - 1987] and autocorrelations [Serio et al., 2000; Masiello et al.,. 2002] of observations with synthetic spectra. The latter finds applications in ...
Philip E. Dennison 1 and Dar A. Roberts 2. 1 Department of Geography and Center for Natural and Technological Hazards,. University of Utah, Salt Lake City, ...
Nov 6, 2017 - Keywords: hyperspectral; image classification; least squares; ... a convolutional neural network is utilized to extract deep features from high ...
AbstractâThis paper presents a new approach for hyperspec- tral image visualization. A bilateral filtering-based approach is presented for hyperspectral image fusion to generate an appro- priate resultant image. The proposed approach retains even t
Schaum and. Stocker (2004) predict what one image should look like at the later time by using methods based on covariance equalization followed by the chronochrome algorithm or matched change detection to ... information is recorded for each tile in
Jan 12, 2017 - tection  . Using multi view-angle images for urban change detection has .... The Dataset C1 images have similar solar angles, while in Datasets C2 and C3 ..... Usually if the coefficient value falls within 2 , Ï Ï. Â± ..... 
Nov 13, 2015 - proposed an HMM-based anomalous signal detection algorithm to ... are used for the manual selection of a validation set. The other dataset are high spatial resolution images from Google Earth, used for examining the validation set. ...
Seeded and non-seeded helicopter gearbox fault results (CH-46E and CH-47D, respectively) show that bispectral algorithms can detect faults at the level of an ...
(e) Touch. (f) Microp1. (g) Microp2. (h) Temp. Figure 4: The phone1 data. is carefully designed to avoid such difficulties which are very likely to occur in dense ...
applied to the analysis of time series (see, e.g., Kim, Cho and. Lee, 2000 in ; Lee, Ha .... non-change point parameters in the regression model is assessed before (and after) a ..... Borror, Montgomery and Runger (1999) in  compared the ...
Working Report 2011-26
Detection of Environmental Change Using Hyperspectral Remote Sensing at Olkiluoto Repository Site Jyrki Tuominen Tarmo Lipping
Olkiluoto FI-27160 EURAJOKI, FINLAND Tel
Fax +358-2-8372 3709
Working Report 2011-26
Detection of Environmental Change Using Hyperspectral Remote Sensing at Olkiluoto Repository Site Jyrki Tuominen Tarmo Lipping Tampere University of Technology Pori Unit
Working Reports contain information on work in progress or pending completion.
The conclusions and viewpoints presented in the report are those of author(s) and do not necessarily coincide with those of Posiva.
ABSTRACT In this report methods related to hyperspectral monitoring of Olkiluoto repository site are described. A short introduction to environmental remote sensing is presented, followed by more detailed description of hyperspectral imaging and a review of applications of hyperspectral remote sensing presented in the literature. The trends of future hyperspectral imaging are discussed exploring the possibilities of long-wave infrared hyperspectral imaging. A detailed description of HYPE08 hyperspectral flight campaign at the Olkiluoto region in 2008 is presented. In addition, related pre-processing and atmospheric correction methods, necessary in monitoring use, and the quality control methods applied, are described. Various change detection methods presented in the literature are described, too. Finally, a system for hyperspectral monitoring is proposed. The system is based on continued hyperspectral airborne flight campaigns and precisely defined data processing procedure. Keywords: hyperspectral imaging, environmental change, change detection, remote sensing
Ympäristönmuutoksen havainnointi hyperspektrisen kaukokartoituksen avulla Olkiluodon loppusijoitusalueella TIIVISTELMÄ Tässä raportissa käsitellään Olkiluodon hyperspektriseen monitorointiin liittyviä menetelmiä. Raportin alussa esitetään lyhyt johdanto ympäristön kaukokartoituksesta. Johdantoa seuraa yksityiskohtainen kuvaus hyperspektrisestä kaukokartoituksesta ja katsaus tieteellisessä kirjallisuudessa esitettyihin hyperspektrikuvauksen sovelluksiin. Raportissa tarkastellaan myös hyperspektrikuvauksen kehitysnäkymiä ja esitetään pitkäaaltoisen infrapunahyperspektrikuvauksen mahdollisuuksia Kesällä 2008 suoritetusta HYPE08 hyperspektrikuvauksesta esitetään yksityiskohtainen kuvaus. Kuvauksen yksityiskohtien lisäksi esitetään käytetyt esi- ja ilmakehäkorjausmenetelmät sekä käsitellään hyperspektrikuvauksen laadunvalvontajärjestelmää. Myös useimmat kirjallisuudessa esitetyt muutoksen ilmaisumenetelmät kuvataan tässä yhteydessä. Raportti päättyy ehdotukseen Olkiluodon hyperspektrisestä monitorointijärjestelmästä. Järjestelmä perustuu toistuviin hyperspektrikuvauksiin ja täsmällisesti määritettyyn kuvausaineiston käsittelyyn. Avainsanat: hyperspektrikuvaus, ympäristömuutos, muutoksen ilmaisu, kaukokartoitus.
TABLE OF CONTENTS ABSTRACT TIIVISTELMÄ TERMS AND ABBREVIATIONS..................................................................................... 3 1
HYPERSPECTRAL CHANGE DETECTION AT OLKILUOTO REPOSITORY SITE .................................................................................................................. 23 3.1 3.2 3.3 3.4 3.5
The causes of environmental change ................................................................. 7 Environmental Remote Sensing ......................................................................... 8 Hyperspectral remote sensing of the environment ........................................... 11 Future trends in hyperspectral imaging ............................................................. 16 Remote sensing in nuclear safeguard applications .......................................... 17
HYPE08 hyperspectral flight campaign ............................................................ 23 Quality control in hyperspectral imaging ........................................................... 26 Pre-processing of hyperspectral data ............................................................... 29 Atmospheric correction of hyperspectral data................................................... 30 Change detection .............................................................................................. 36 PROPOSAL FOR HYPERSPECTRAL MONITORING OF OLKILUOTO REPOSITORY SITE ......................................................................................... 39
4.1 Hyperspectral monitoring system...................................................................... 39 4.2 Integration with other environmental data ......................................................... 40 5
Hyperspectral imaging offers unique opportunities in the environmental monitoring of Olkiluoto nuclear power plant and repository site: compared to environmental surveys based on field sampling the spatial coverage and the density of samples are tremendously better. Sampling distance is just a few meters while the study area can easily exceed 1000 square kilometers (the nuclear site is, of course, smaller but for data acquisition for the safety case of the repository, a wider area is required (Haapanen et al. 2009)). Each recorded pixel contains detailed quantitative and qualitative environmental information. Although the potential of hyperspectral remote sensing is great, there are special issues that arise when this unique type of data is concerned. For example, the processing of hyperspectral data involves many sophisticated correction algorithms. The validation process of corrected data must be profound and it should be done using accurate field measurements. There are some issues in data processing where more research must be done in order to develop methods most suitable for the monitoring of the Olkiluoto site. Posiva Oy has considered the use of hyperspectral monitoring for a few years. In 2004 a large regional hyperspectral flight campaign HyperGeos was planned. This campaign, planned to cover the whole Satakunta province, was coordinated by Geological Survey of Finland. Unfortunately the plan was not realized. However, the careful planning laid ground to a new flight campaign: in June 2008 the hyperspectral flight campaign HYPE08 was conducted. The campaign was a joint effort of Posiva Oy, Tampere University of Technology Pori unit and the city of Pori. The Pori unit started the research of hyperspectral data processing in 2006 in close co-operation with Geological Survey of Finland. Partly funded by Posiva, the signal processing group of Pori unit has developed methods for atmospheric correction, quality control and change detection described later in this report. The HYPE08 flight campaign successfully produced high-quality baseline data. Most of the initial problems related to data processing have been solved but the research work is to be continued to further develop change detection methods as well as the manual interpretation procedure. Another hyperspectral flight campaign should be arranged in order to record a second data set to be used in change detection. Lessons learned during the HYPE08 campaign allows to develop more efficient field work during the flights. Development of a hyperspetral monitoring system is a continuous R&D process where the most significant results are shown only after the work of several years. A map of the acquisition areas of HYPE08 is shown in Figure 1.1. In addition to area 3 shown in Figure 3.1 areas 1 and 2, located north of Olkiluoto, were imaged during the HYPE08 campaign. The imaging of those areas was funded by Tampere University of Technology Pori Unit.
Figure 1.1. A map showing the acquisition areas of HYPE08 in South West Finland. Area 3 was commissioned by Posiva Oy. (Map with permission from Land Survey of Finland 952/MML/10.)
REMOTE SENSING OF ENVIRONMENTAL CHANGE
There has been a continuous environmental change on the Earth throughout the history. The extent of this change varies from minor to extreme events. Desertification and ice ages are examples of extreme environmental changes. The state of the environment today in Scandinavia has totally changed after the last ice age occurred a little over 10 000 years ago, and the development continues as post-glacial rebound of the bedrock. Monitoring of environmental changes is essential for a nuclear waste repository site both in terms of monitoring the lack of environmental effects as well as to produce reliable data and time series to feed in the safety assessments where time spans are typically from days (operational) to several millennia (long term dose assessment). Remote sensing techniques offer means to collect information e.g. on land use, vegetation quality, and quantity useful also in these respects.
The causes of environmental change
The global climate is changing. Increased temperatures and levels of atmospheric carbon dioxide as well as changes in precipitation and in the frequency and severity of extreme climatic events are just some of the changes occurring (Moore & Allard 2008). These changes are having notable impacts on ecosystems. Climate change has both direct and indirect impacts on the environment. Climate, temperature and precipitation in particular, have a very strong direct impact on the development, reproduction and survival of different plant and animal species. Drought is one of the major indirect impacts of climate change. It is one of the most important climate-related events through which rapid ecosystem changes can occur as it affects the survival of existing animal and plant populations (Gordo & Sanz 2006). Phenology is the timing of seasonal activities of plants and animals such as flowering or breeding. Since it is in many cases temperature dependent, phenology can be expected to be influenced by climate change. All organisms modify their environment, and humans are no exception. As the human population has grown and the power of technology has expanded, the scope and nature of this modification has changed drastically (Vitousek et al. 1997). The term “humandominated ecosystems” applies nearly to all of Earth. The following human actions are the major causes of human-induced environmental changes: -
energy production mining manufacturing transport agriculture settlements tourism recreation
The production of energy has a significant contribution to the emissions of greenhouse gasses and air pollutants, which are causing environmental changes. Environmental hazards due to energy production like oil spills in off-shore oil drilling are another cause of environmental changes. Mining causes direct landscape changes and in many cases enables the emission of hazardous substances into the environment. The manufacturing of commodities has environmental impacts in all production phases; obtaining the raw materials, waste produced in the manufacturing process and emissions into the air and water areas cause environmental changes. Transportation is the major source of air pollution. Hazards during the transportation are also changing the environment. Oil spill from Exxon Valdez vessel in 1989 caused drastic environmental changes over a wide geographic area (Peterson et al. 2003). Environmental changes due to agriculture have increased rapidly during the past century. Cultivation of forests into fields has significant impacts on the carbon cycle of the Earth. Settlements always alter the ecosystem of the surrounding area and they also tend to create waste, most easily dumped nearby. Tourism tends to increase littering and traffic, which causes environmental changes. Even recreational activities of humans can cause environmental changes. For example excessive sport fishing can lead to extinction of certain species. There are always changes in ecosystems, even without human activities or change in climate. In general these natural changes are slow, but there are natural hazards that cause rapid and drastic environmental changes like volcanic eruptions, floods and earthquakes. Monitoring of these natural environmental changes is becoming extremely difficult because most of the Earth belongs to the human-dominated ecosystem leaving only some national parks and other preservation areas for study sites and in many cases even those areas have experienced some influence of humans in the past.
Environmental Remote Sensing
Remote sensing can be defined as the science of observation from a distance. Thus it is contrasted to in situ sensing, in which measuring devices are either immersed in, or at least touch the object(s) of observation (Barret & Curtis 1992). The use of remote sensing does not exclude or make more difficult the use of in situ measurements. In many cases the best result can be obtained by using remote sensing data together with in situ measurements. The latter can be used as a reference in the calibration of remote sensing algorithms, and the results of remote sensing can be validated using in situ measurements. Today, remote sensing is a key technology for environmental monitoring. The applications of environmental remote sensing are numerous: -
Geology Hydrology Precision agriculture Forest health and inventory Water quality Land use mapping Ice and snow cover mapping Etc.
The apparatus currently used for remote sensing can be divided into two groups: active and passive systems. The active sensors generate and transmit a signal toward the target, and receive and record the returned signal after its interaction with the target. The relationship between the transmitted and received signal is used to characterize the condition or state of the target. Active microwave sensors (radars) and LiDARs (Light detection and ranging) belong to that category. The passive sensors do not generate or transmit signals, however; they detect and record the natural electromagnetic energy reflected and/or emitted from the target. The magnitude and shape of the signal are indicators of the condition/state of the target. Cameras, radiometers, scanners and spectrometers are examples of this category (Asrar 1989). The following data acquisition methods are frequently used in environmental remote sensing: - Aerial photographs, black and white or colored - False color - Multispectral images - Hyperspectral images - Synthetic aperature radar (SAR) images - LiDAR (Light detection and ranging) Aerial photographs are just like conventional photographs but they are taken from an aircraft or a satellite using a high quality camera. A false color image is a composite of color photograph and near infrared (NIR) photograph which is built out of the colors yellow-green and red of the visible spectrum and the red of the near infrared spectrum. All green vegetation has a very high reflectance at NIR region and therefore false color images are very useful in the detection of vegetation. In Figure 2.1 an aerial color image and a false color photograph of the same urban residential area is shown. From the false color photograph it can be seen how all green vegetation is highlighted with red color. Multispectral remote sensing involves the acquisition of visible, near infra-red, and short-wave infrared (SWIR) images in several broad wavelength bands.
Figure 2.1. Aerial color (left) and false color (right) photographs of a urban residential area. (Photos on the courtesy of the city of Pori.) Different materials reflect and absorb differently at different wavelengths. Therefore, it is possible to differentiate among materials by their spectral reflectance signatures as observed in these remotely sensed images, whereas direct identification is usually not possible. Hyperspectral imaging systems acquire data in over one hundred contiguous spectral bands. While multispectral imagery is useful in discriminating land surface features and landscape patterns, hyperspectral imagery allows for identification and characterization of materials. SAR images have the advantage of being not affected by cloud cover which often prevents the use of optical remote sensing especially in northern latitudes. SAR images are often used in ice and snow cover monitoring. LiDAR is mostly used to generate digital elevation models (DEM) while its intensity component can also be used in land use classification. The most common characteristics of different remote sensing sensors are the spatial, temporal and spectral resolution. Spatial resolution specifies the pixel size of a remotely sensed image covering the Earth’s surface. Temporal resolution specifies how often a target can be imaged. Time series analysis is an important technique in environmental monitoring and it requires new remote sensing images at certain intervals, e.g., a new image per season. Sensor’s spectral resolution specifies the number of spectral bands in which the sensor can collect reflected radiance. However, the number of bands is not the only important aspect of spectral resolution. The position of bands in the electromagnetic spectrum and the bandwidth of channels are important, too. The use of remote sensing in environmental monitoring has some advantages and disadvantages compared to conventional in situ measurements: The main advantages are: - good spatial coverage - high spatial resolution
cost effectiveness access to remote and/or restricted areas efficiency of data acquisition ability to capture information not visible to the eye
whereas the main disadvantages are: - difficulties in data interpretation - the need for calibration over different acquisition areas - the need for complicated correction procedures i.e. radiometric, geometric and atmospheric corrections - susceptibility to weather conditions
Hyperspectral remote sensing of the environment
Hyperspectral remote sensing is, in some ways, the ultimate optical remote sensing technology. It seeks to uniquely identify and map surface materials of the Earth through the measurement of continuous relatively high-resolution spectra of each pixel in spatially high-resolution images of the surface (McDonald et al. 2009). Hyperspectral data can be used for the mostly unambiguous direct and indirect identification of surface materials and atmospheric trace gases, measurement of their relative concentrations, and subsequently assignment of the proportional contribution of mixed pixel signals (e.g. spectral unmixing), derivation of their spatial distribution (e.g. mapping), and finally their evolution over time (multi-temporal analysis). The sensors used in hyperspectral data acquisition are called imaging spectrometers. The operation of a modern imaging spectrometer is based on the push-broom principle, meaning that it builds the image one line at a time. The most important characteristics of imaging spectrometers are: - spectral range - spectral resolution - spatial resolution - signal-to-noise ratio Earth observation based on imaging spectroscopy has been transformed in little more than two decades from a sparsely available research tool into a commodity product available to a broad user community (Schaepman et al. 2009). This is mainly due to remarkable advances in sensor technology. When the first AISA imaging spectrometer manufactured by Specim Ltd. was introduced in 1995, it was able to record hyperspectral data in 17 wavelength channels. Today the latest spectrometer model by Specim Ltd., AISA Dual, allows data acquisition in 492 wavelength channels. Signalto-noise ratio (SNR) is another characteristic where tremendous progress has been achieved in sensor development. A good example of this kind of advancement is the AVIRIS spectrometer developed by the NASA Jet Propulsion Laboratory. In the early 1990s its signal-to-noise ratio could be measured in the tens to hundreds depending
upon the wavelength region. Since then, AVIRIS has evolved through major upgrades and constant hardware and software adjustment, to provide spectra with effective signalto-noise performances of several hundreds to thousands (Asner 2008). The payoff in sensor performance is clearly seen in the spectra and in derived products such as nitrogen concentration maps of vegetation. Today, researchers are able to apply consistent methodologies to hyperspectral analysis in the field of remote chemistry, whereas previous studies with older AVIRIS technology struggled to do so. This kind of evolvement has been achieved with other sensors as well. Earlier it was common that the spectral range of hyperspectral sensors was limited to the visible and infra-red regions of the electromagnetic spectrum. Today, all commonly used sensors, i.e., AVIRIS, AISA, HyMap and CASI, allow data acquisition capturing the spectra from visible to short-wave infrared region (400-2500 nm). Advances in sensor technology have also enabled the use of better spatial resolution in data acquisition. The acquisition time and cost needed to obtain data with certain spatial resolution is considerably less than it used to be. While the basic concept behind hyperspectral imaging is relatively straightforward, the implementation is far more difficult. In most applications hyperspectral sensors collect sunlight reflected from a target area into a number of narrow wavelength bins, usually across the visible, near infrared and shortwave infrared portions of the electromagnetic spectrum. These bins of photons are both spectrally and spatially resolved and thus can be used to generate a variety of “images” of the target area (Pabich 2002). The key to hyperspectral imaging lies in the concept of spectral signatures. In simple terms, all materials transmit, reflect or absorb electromagnetic radiation based on the inherent physical structure, chemical composition of the material and the wavelength of the radiation. To put it in another way, for any given material the amount of electromagnetic radiation that is absorbed, reflected or transmitted varies with the wavelength of the radiation. Spectral signatures of three different vegetation species measured using field spectrometer near the Olkiluoto site are shown in Figure 2.2.
Figure 2.2. Spectral signatures of grey lichen, brown lichen and grass measured near the Olkiluoto site using a field spectrometer. If the reflectance values for a given material are plotted across a range of wavelengths, the resulting curve is referred to as the spectral signature of that material. Because the spectral signature is different and indeed unique for each material, it is possible to discriminate between materials based on the differences in the spectral signatures of the materials. For example, hyperspectral imaging can differentiate between lichen and grass. Not only are broad differences such as those just noted detectable, it is also possible to identify particular materials such as different lichen species based on a comparison against a database of known signatures. For example, as shown in Figure 2.2, in a hyperspectral image from an archipelago close to Olkiluoto it should be possible to differentiate between lichen species whose spectral signatures are shown. This concept forms the basis of hyperspectral imaging. Hyperspectral imaging has enabled applications in a wide variety of environmental studies. Some of the pioneering research papers are mentioned here but the complete list of applications is beyond the scope of the present work. The prime motivation for the development of imaging spectrometry was mineralogical mapping of surface soils and rock outcrops (Goetz 2009). The reflectance spectra of minerals are rich in detail especially in the short-wave infrared (SWIR) region. Detailed spectral features make it possible to identify minerals using imaging spectrometry in a reliable manner (Goetz & Srivastava 1985). Hyperspectral identification of arid and semiarid soils has also been successful despite the problems related to changing moisture content (Ben-Dor & Banin 1994). Maps of expansive clay soils, important in construction engineering, can also be created using hyperspectral data (Chabrillat et al. 2002).
Generally, only approximately 30 % of the land surface is relatively free from vegetation and the remaining 70 % is covered by vegetation to the extent that the soil is rendered inaccessible to remote sensing identification (Siegal & Goetz 1977). Hyperspectral imaging is usually restricted to study the Earth’s surface, but some applications have successfully penetrated below the surface by utilizing the indirect features in the surface caused by buried objects. Detection of landmines and applications in archeology are good examples of indirect hyperspectral identification of buried objects and materials (Bowman et al. 1998; Cavalli et al. 2007). Oil and mineral exploration are the most important commercial applications of hyperspectral imagery. The spectral characteristics of oil seeps and oil-impacted soils are generally too subtle to be detected by traditional multispectral sensors, but the high spectral resolution of hyperspectral data provides adequate means to identify these targets (Ellis et al. 2001). Mineral exploration has also benefited greatly from the use of hyperspectral imagery (Kruse et al. 1993). Knowledge of the distribution of vegetation on the landscape can be used to investigate the ecosystem. Numerous vegetation mapping applications using hyperspectral data have been published in the literature. Recent studies have indicated that sophisticated identification methods like spectral feature analysis, e.g., can provide reliable results in vegetation mapping even in challenging environments (Kokaly et al. 2003). Tree species classification using hyperspectral data has also been the subject of ongoing research for a few decades. While most approaches employ conventional spectral classification methods, Wessman et al. (1988) were the first to attempt the identification of tree species based on nitrogen and lignin content in the foliage. Hyperspectral analysis of vegetation health and vigor is usually done by means of Vegetation Indices (VI's). VI's are combinations of surface reflectances at two or more wavelengths designed to highlight a particular property of vegetation. Most commonly studied properties of vegetation are chlorophyll, water, leaf pigment and carbon content as well as light use efficiency (Tuominen et al. 2009). VI maps are used to study the deterioration of forest health due to contamination. The spatial distribution of two VI’s: normalized difference vegetation index (NDVI) and Anthocyanin Reflectance Index (ARI) around the Lahnaslampi talc mine in northeast Finland is shown in Figure 2.3. The most advanced form of hyperspectral vegetation analysis is remote chemistry, and even though it is still partly in an experimental phase, very promising results have been published. In forestry applications foliar chemistry related variables such as lignin, carbon and nitrogen content have been estimated with good results (Curran 1989). Precision agriculture is another potential area of hyperspectral remote chemistry; Monteiro et al. (2007) presented a method for estimating amino acid content in soybean crops.
Figure 2.3. Spatial distribution of NDVI and ARI vegetation indices around Lahnaslampi talc mine in Northeast Finland (Tuominen et al. 2009). While open oceans can be studied using multispectral imagers such as SeaWiFS, MERIS and MODIS, studies of the coastal zone benefit more from hyperspectral imaging, which makes it possible to estimate important water quality parameters such as chlorophyll, colored dissolved organic matter (CDOM) and suspended inorganic matter contents as well as turbidity (Dekker, 1993). In shallow and clear waters it is also possible to identify submerged vegetation and the type of the sea bottom through the water column (Williams et al. 2003). Hyperspectral imaging is equally applicable to studying the properties of ice and snow, in particular the grain size (Nolin & Dozier 1993). The use of hyperspectral imagery for contamination mapping has proven valuable when ensuring ecosystem integrity and protecting human health against contaminated sites. The use of hyperspectral data allows detection of lesser quantities of hazardous materials and more accurate identification of materials compared to other remote sensing methods. Swayze et al. (2000) estimated savings of millions of dollars in the cleanup of the Leadville Superfund Site in which AVIRIS hyperspectral images combined with field spectral measurements indentified the waste piles with the greatest potential for leaching heavy metals into streams and groundwater. Several studies have showed the potential of hyperspectral imagery in the detection of toxic substances such as selenium (Mars & Crowley 2002). Hyperspectral contamination mapping is not restricted to land areas, as it has been used successfully to study offshore water areas as well. Oil spills and other discharges of hazardous chemicals can be identified using hyperspectral imagery (Salem & Kafatos 2001).
Future trends in hyperspectral imaging
Advancements in sensor technology will very likely continue. There are some practical limitations, however. Imaging spectrometers based on silicon charge-coupled device (CDD) arrays can be built relatively inexpensively because the arrays and supporting electronics have other commercial uses and the prices are concomitantly low (Goetz, 2009). On the other hand, hyperspectral imagers operating beyond 1 μm wavelength require much more expensive detectors, which are manufactured in low volume. Therefore full-range imaging spectrometer systems will continue to be rather expensive, which will show in data acquisition costs. An ambitious step in the development of hyperspectral sensors will be the operation of ultraspectral instruments with spectral bandwidth smaller than 1 nm and number of channels around 1000, enabling trace gas detection and quantification (Richter 2005). Trace gasses such as ozone have very fine spectral features. The identification and quantification of these gasses requires sensor technology with very high spectral resolution capabilities. Hyperspectral image processing is very likely to benefit from advances in algorithm development. Currently, there are only a handful of university programs teaching remote sensing involving advanced techniques such a hyperspectral imaging. The number of universities and other research institutions participating in the research and development of hyperspectral image processing is likely to increase, which will raise the use of new advanced methods. Continuing advances in computing technology offer access to computing platforms capable to implement complicated algorithms and handle large volumes of data. One way to reduce data acquisition costs is to use Unmanned Aerial Vehicles (UAV) as platforms for hyperspectral sensors. There are several sensors currently available which are designed for UAV use. Technical Research Centre of Finland (VTT) has developed a new miniaturized hyperspectral sensor. The sensor weights only 350 grams, which allows the use of light UAV platform. The operational wavelength range of the prototypes built so far can be tuned in the range 400-1100 nm and the spectral resolution is in the range of 5-10 nm (Saari et al., 2009). Presently the spatial resolution of the sensor is 480 x 750 pixels. The use of light UAV makes possible to organize simplified, cost effective flight campaigns. Currently available UAV sensors operate in the VNIR range, but the emergence of full-range (400-2500 nm) sensors can be expected. In the future, organizations using hyperspectral imaging can operate their own UAV sensors and planes, allowing flexible and frequent data acquisition. There is also a need for a hyperspectral imager in orbit that can produce images of the quality and resolution of airborne sensors and that is radiometrically stable and pointable (Goetz, 2009). The high cost of airborne hyperspectral campaign usually prohibits seasonal time-series studies. Given the dynamic character of vegetation cover, a snapshot in time is not nearly as revealing as a time sequence. The availability of lowcost satellite-borne hyperspectral data acquisition will eventually make possible such studies, but it can easily take a decade or two. Except for the experimental Hyperion sensor, no commercially available hyperspectral data are currently available from the orbit. However, several countries are planning space-borne hyperspectral missions. Some missions have been stalled but German EnMAP mission looks very promising.
The EnMAP sensor should measure in the 420-2450 nm spectral range at a varying sampling of 6.5-10 nm. Approximate ground sampling distance is 30 meters and specified signal-to-noise ratios are comparable to current airborne sensors (Guanter et al. 2009). EnMAP is scheduled to launch in 2013. The development of airborne LWIR hyperspectral imagers will probably be the most significant advancement in hyperspectral technology. The use of the LWIR range (8-14 um) enables remote identification of chemicals, radioactive materials and gases in many cases where current full-range (400-2500 nm) hyperspectral imagery fails. The use of LWIR has two benefits: atmospheric attenuation is low and many materials have distinct spectral features in the LWIR range (i.e. absorption points and reflectance peaks). But, unfortunately, hyperspectral imaging in the LWIR region is still in an early stage of development. This is true especially in airborne remote sensing, but also in industrial processes or quality control and laboratory studies (Holma et al. 2009). There are very few instruments available which can provide both feasible performance and usability. Spatially Enhanced Broadband Array Spectrograph System (SEBASS) and Airborne Hyperspectral Imager (AHI) instruments are most widely used pushbroom imagers in airborne LWIR experiments (Kirkland et al. 2002; Mares et al. 2004). The use of the SEBASS sensor has produced promising results in geological mapping and plant species identification (Vaughan et al. 2003; da Luz & Crowley 2010). AHI has been used successfully for a wide range of applications, including mine detection, gas detection and geological, urban, and military mapping applications (Lucey et al. 2001). Both instruments are bulky and require intensive care and maintenance during operation. Specim Ltd. has introduced a new airborne LWIR sensor, AisaOWL, which has overcome these operational restrictions. According to Holma et al. AisaOWL is capable to record LWIR data with high spatial and spectral resolution as well as excellent signal-to-noise characteristics (Holma et al. 2009). AisaOWL is clearly a step towards commercial LWIR data acquisition from the scientific experiment phase. Although the research on airborne LWIR imaging is still at its early stage, the potential in chemical, gas and radioactive material identification can be evaluated by studying the results obtained using FTIR and chromotomographic imaging spectrometers in the field and in laboratories. Gittins & Marinelli (1998) published results on successful LWIR identification of hazardous gasses. LWIR hyperspectral imaging is also capable in identifying chemicals used in chemical warfare (Farley et al, 2006). FTIR imaging has been successfully used to identify radioactive materials (Puckrin & Theriault 2004). 2.5
Remote sensing in nuclear safeguard applications
Nuclear safeguards are measures to verify that states committed to non-proliferation of nuclear weapons comply with their international obligations not to use nuclear materials (plutonium, uranium and thorium) for nuclear explosives purposes. Nowadays, nuclear safeguards staff of International Atomic Energy Agency (IAEA) and other authorities have access to a growing body of remote sensing imagery that provides a range of types of information. High-resolution panchromatic imaging that provides detailed information by photo-interpretation is still the most commonly used modality,
but multispectral, thermal and hyperspectral imaging are also applied (Borstad et al. 2007). Published research results in the field of nuclear safeguards quite clearly show that efficient remote sensing based monitoring system covering the whole processing chain of nuclear material is a feasible and effective tool in the verification. In nuclear energy production, the processing chain starts from the uranium mine and ends up at the final repository site. The repository site differs from other processing sites in terms of its timeline: the time that nuclear material is stored at the site far exceeds the storage times in other processing sites as the nuclear materials have very long, millennial, half lives. Remote sensing technology offers clear advantages in long-term monitoring of a repository site; cost effectiveness and spatial coverage. With adequate field verification and calibration measures the quality and accuracy of remote sensing monitoring can be guaranteed. Remote sensing applications for the monitoring of operational nuclear plants have generally been limited to studying the thermal effects of cooling-water discharge. Chen et al. (2003) presented a method based on local algorithm using Landsat-5 Thematic Mapper thermal band data to estimate water temperature. The presented technique provides an effective means to quantitatively monitor the intensity of thermal discharge and to retrieve a very detailed distribution pattern of thermal effects of discharge. Borstad et al. (2001) presented a potential application of satellite imagery for international safeguards using limnological and optical knowledge to detect discharges from nuclear facilities. Most nuclear plants are located either on the shoreline of seas or lakes or at the connecting channels (e.g. river) because of the large volumes of water needed for cooling and condensing steam in the power generation process. These water discharges provide several potential mechanisms to monitor the activities of these power plants using remote sensing techniques. They may produce several kinds of local hydrological changes in the receiving water bodies, including: -
altered thermal structure in the vicinity of the discharge altered current patterns at intake and discharge structures altered surface wave patterns alteration of the concentration of suspended inorganic material scouring caused by increased flow near intake and discharges stimulation of phytoplankton and/or benthic algal growth altered salinity gradients in estuaries
All of these phenomena are potentially observable using various remote sensing techniques. Häme (2003) proposed the use of satellite-borne remote sensing for the safeguard of a nuclear waste repository site. The main purpose of the proposed monitoring system is to prevent the illegal use of repository site and detect unauthorized activities such as quarrying around the repository or building of undeclared underground facilities. Proposed monitoring system consists of baseline data acquisition, routine monitoring and an optional alarm survey. Space-borne Synthetic Aperature Radar (SAR) is suggested as the principal instrument to collect the monitoring data because of its allweather capability: the SAR technology provides means for monitoring day and night
despite of the snow, clouds, rain or lack of light. Methods as coherent change detection, interferometry and polarimetry help to extract information on the surface cover (i.e. infrastructure of the site) as well as on surface and terrain heights, surface movements and deformations due to drilling, mining or camouflage (Niemeyer 2009). In addition to SAR imagery, optical and thermal remote sensing data is used in order to overcome the limitations of moderate spatial resolution of the SAR data. Uranium mines are often located in remote areas and are difficult or expensive to access (Truong et al. 2003). There has been much interest in using multispectral and hyperspectral sensors for remote verification of uranium mining and milling. The use of these techniques by international monitoring communities has been seen as a way to minimize costs while improving inspection performance (Smartt et al. 2005). Geology was one of the first disciplines to benefit from hyperspectral remote sensing. Therefore, most of the theory is developed relative to earth material interactions and it provides good opportunities for geological remote sensing. Successful identification of a material using hyperspectral data is highly dependent on the size of the area covered predominantly by that particular material. Geologic formations tend to be large, meaning that there is a large amount of pure pixels which represent only one material. Monitoring of uranium mines using remote sensing is far from being a routine procedure. In the case of uranium mining activities, the abundance of uranium in the ore can be 0.3 % and significantly less in the tailings. Even under optimal circumstances, using currently available airborne hyperspectral sensors to spectrally detect uranium in the ore will be challenging due to uranium’s very low concentrations (Smartt et al. 2005). The use of airborne hyperspectral imaging in the monitoring of uranium sites have been addressed in several research papers. Neville et al. (2001) used hyperspectral data and identified spectral signatures typical to uranium mine with moderate results. Levesque et al. (2001) used imagery from Probe-1 hyperspectral sensor covering the Pronto mine tailings near Elliot Lake, Ontario, Canada. The prime research task was to determine if uranium mine tailings could be distinguished from other types of mine tailings using unique mineral compound absorption features. While uranium tailings could be distinguished from copper tailings, the absorption features of uranium dioxide could not be identified in the extracted end-members. This is quite expected considering the low concentration of uranium (0.1 % in this study). Airborne hyperspectral sensors offer superior data quality compared to satellite-borne sensors, but in many areas the acquisition of airborne data is simply not feasible due to practical or administrational restrictions. Often these cases are the most interesting ones from the safeguard monitoring point of view. Therefore, it would be highly desirable to be able to use satellite-borne data. Borstad et al. (2006) presented a study where the feasibility of material identification and tracking using satellite-borne hyperspectral data for nuclear safeguard monitoring were investigated. The study utilized data collected by the Hyperion sensor onboard EO-1 satellite from the Al Qaim fertilizer plant and the Akashat phosphate mine in western Iraq. The map of the study site is shown in Figure 2.4.
Figure 2.4. Map of Western Iraq. Al Qaim fertilizer plant and Akashat mine are circled red (Borstad et al. 2006). Uranium is present in low concentrations in the Akashat mining area and it was used for extraction from late 1984 to late 1990. The main goal of the study was to determine if it would be possible to monitor the transfer of ore from mine to a processing facility using satellite hyperspectral imagery. The results were encouraging, despite the difficulties in atmospheric correction; the results clearly showed that in order to identify minerals and to track material from one scene to another, it is necessary to use well-calibrated imagery, in which all effects associated with the remote measurement (atmospheric absorption, scattering and viewing geometry) have been removed or compensated for. Only then will it be possible to derive spectral signatures identical to those acquired by ground measurements. The main purpose of the hyperspectral data acquisition campaign of the Olkiluoto repository site in 2008 (HYPE08) was to gather environmental information to be used in the biosphere assessment (Haapanen et al. 2009, Hjerpe et al. 2010) as part of the longterm safety case. Hyperspectral imaging could also be used in the monitoring of radioactive contamination, if there were any except the global background and Chernobyl fallout, by studying the changes in vegetation. The effect of the radioactive contamination on vegetation is twofold: (1) internal radiation and radiotoxicity as biologically available radionuclides are incorporated in the plant tissue through root uptake (Papastefanou et al. 1999); (2) external irradiation from radionuclide concentrations in the soil (Absalom et al. 2001). However, the latter would require rather intensive contamination likely detected also by other means. Concerning the former effect, radioactive contamination changes the rate of photosynthesis and concentration of pigments in the plant leaves (Lichtenthaler 1996). These changes can be seen in the overall light reflectance spectrum of the plant and can be detected using remote sensing techniques.
Figure 2.5. Map of the sampling sites around the Chernobyl nuclear power plant together with corresponding approximate contamination levels (Davids & Tyler 2003). In Eurajoki, the municipality around the Olkiluoto nuclear site, the radiation level is around 0.17 μSv/h (www.stuk.fi). Davids & Tyler (2003) presented results of laboratory experimens and in situ spectroradiometry measurements of silver birch (Bentula pendula) and Scots pine (Pinus sylvestris) across a range of contamination levels in the Chernobyl exclusion zone. The results were used to evaluate whether vegetation stress caused by radionuclide contamination can be detected using remote sensing techniques and whether this stress may be distinguished from vegetation stress related to variation in moisture conditions. Five different Vegetation Indices were tested. Results showed that Chlorophyll Red Edge and the Three Channel Vegetation Index (TCHVI) correlate well with activity concentrations of Sr-90 and Cs-137 in leaves, gamma dose rates and Cs137 inventories in soil. Results also show that both indices are independent of soil moisture, which indicates that contamination-induced stress can be distinguished from soil-moisture related stress. The map of the sampling sites around the Chernobyl nuclear power plant is shown in Figure 2.5. The environment around the Olkiluoto site is dominated by forest and coastal water areas. Monitoring of phytoplankton and benthic vegetation in water area using hyperspectral remote sensing is also a potential safeguard application, especially to the
power plant, but also to the drainage waters of the repository tunnels. Radioactive contamination affects submerged vegetation in the same manner as trees, where radionuclides are incorporated in the plant tissue through water uptake (Papastefanou et al. 1999). From a safeguard standpoint, the time scales over which phytoplankton and benthic plants respond to environmental changes are quite different. Because of their rapid growth rates and short life cycles, phytoplankton responds quickly (Borstad et al. 2001). Therefore remotely detected changes in phytoplankton regime can be indicative of recent radioactive contamination, although it can be difficult to distinguish changes related to radioactive contamination from other environmental changes. Benthic plants on the other hand respond more slowly and would reflect longer term changes. The use of algae and benthic plants as bioindicators for monitoring radionuclide concentrations have been published (Burger 2007) and they have been used in the conventional in situ sampling (e.g. Ilus 2009). In order to determine the feasibility of detecting changes induced by radioactive contamination in submerged vegetation using hyperspectral remote sensing, further research is needed. The development of new hyperspectral sensors operating in the Long Wave Infrared (LWIR) region will raise the nuclear safeguard applications of remote sensing to a new level. Many radiological materials have distinct spectral signatures in LWIR range (8-14 μm), which makes possible to identify these materials using remote sensing. Puckrin & Theriault (2004) have published result of a study where passive standoff detection by Fourier-transform infrared (FTIR) radiometry was used to identify radiological materials such as SrO, I2O5, ThO2 and La2O3. Later Puckrin et al. (2006) published study results in which successful identification of uranium oxides UO2, UO3 and U3O8 was reported. Identification was done using FTIR sensor at standoff distances of 10-60 m. By studying the series of simulations using the radiative transfer model MODRAN4, Puckrin et al. showed that these materials have a high potential of being detected from altitudes up to 1 km above the Earth’s surface.
HYPERSPECTRAL CHANGE DETECTION AT OLKILUOTO REPOSITORY SITE
HYPE08 hyperspectral flight campaign
The HYPE08 flight campaign was carried out in July 2008. The main goal of the hyperspectral data acquisition was to capture the hyperspectral baseline of environment around Olkiluoto repository site before the nuclear construction license. Baseline data is the reference used in change detection when the data of future campaigns will be analyzed. Baseline data contains the information on each ground pixel's surface material and its state. The planning phase preceding the actual flight campaign included calculation of flight lines, search and verification of sites used for field measurements and co-timing of operations.
Figure 3.1. Flight lines of the HYPE08 campaign in the Olkiluoto-Rauma area commissioned by Posiva Oy. (Map with permission from Land Survey of Finland 952/MML/10.)
The flight campaign was accomplished in collaboration with the following partners: -
Aero Media Ltd., Hungary, flight operations University of Debrecen, Hungary, data acquisition , pre-processing of data Pöyry Environment Ltd., planning and coordination Geological Survey of Finland, field measurements using portable spectrometer Luode Consulting Ltd., water quality measurements in field campaign Tampere University of Technology, Pori, planning, research, data postprocessing Posiva Oy, customer, field work
The size of the imaged area is over 600 km². Half of the imagery covers coastal water areas. The rest of the data covers rural forest and agriculture areas as well as urban area around the city of Rauma. The map showing the actual flightlines in Olkiluoto-Rauma area is presented in Figure 3.1. In addition to the area shown in Figure 3.1, areas located north of Olkiluoto were imaged during the HYPE08 campaign. The imaging of those areas was funded by Tampere University of Technology Pori Unit. The start of the flight campaign and the selection of flights days were planned based on the information provided by the commercial weather forecast service Foreca Ltd. Despite of careful planning, the campaign suffered from severe cloud cover in many days. The total number of recorded flight lines was 27, of which 23 flight lines were recorded on 4th of July and 4 flight lines 13th of July. The cloud cover on both days was absent providing homogenous solar irradiation from ground surface. The acquisition of 4 pre-planned flight lines was cancelled due to excessive cloud cover. The flight altitude during the acquisition was 1.9 km leading to ground resolution of 2.5*2.5 m² per pixel. The acquisition was done using Piper Pa23-250 aircraft carrying an AISA dual imaging spectrometer owned and operated by the University of Debrecen, Hungary. The AISA dual spectrometer collects reflected solar radiation in 481 bands from 399 to 2452 nm wavelength. This includes the visible, near infrared and shortwave infrared regions of the electromagnetic spectrum. The spectral resolution is 3.3 nm at VNIR range and 12 nm at SWIR range.
Figure 3.2 Jukka Laitinen (Geological Survey of Finland) making solar irradiation measurement using portable spectrometer during the field campaign. (Photo by Arto Vuorela.) A field campaign was conducted during the flights. The purpose of the field campaign was to collect data for the validation and calibration process of the hyperspectral data. Geological Survey of Finland made spectral measurements at REF and PIF targets using ASD FieldSpec Pro portable spectrometer. Measurement of solar irradiation using portable spectrometer is shown in Figure 3.2. REF targets are spectrally homogenous tarpaulins laid to the ground during the overflight. Pseudo Invariant Feature (PIF) targets are natural homogenous areas on the ground surface. The field teams were moving along the flight lines following the schedule of the aircraft. There was no direct communication link between field teams and aircraft. The flight control was able to relay some messages, but the timing of the field work would have been easier with better communication arrangements.
Figure 3.3. The route of water quality measurements around Olkiluoto Island. The water sample collection points are marked using red dots and the reference sample points using green crosses (Lindfors et al. 2008). As part of the field campaign, Luode Consulting Ltd. conducted on Posiva's commission water quality measurements using optical flow through instrument installed in to the boat (Lindfors et al. 2008). The purpose of these measurements was to collect data to be used in the validation of water quality maps derived from hyperspectral data. The flow-through instrument recorded continuously the temperature, salinity, turbidity and chlorophyll content along the route (Figure 3.3). Discrete measurements of nitrate and suspended organic carbon were also made (Lindfors et al. 2008). Also discrete water samples from ditches around Olkiluoto Island were collected and measured using the same flow-through instrument. 3.2
Quality control in hyperspectral imaging
Hyperspectral change detection analysis requires the use of at least two different datasets. Unavoidably, the acquisition of datasets has been made under different circumstances. Numerous factors affecting the data may have been changed, e.g., sensor type, weather conditions and flight time. All data used in the change detection should be submitted to quality control procedure in order to ensure the reliability of results, even though the issue of quality control has not been addressed widely in the literature.
Published quality-related studies have concentrated on evaluating the degradation of the hyperspectral data in the compression process (Penna et al. 2006). In general, quality in hyperspectral imaging process means that every phase in the acquisition and data processing chain is defined and documented. The quality of hypespectral data can be inspected by evaluating three different quality issues: -
Quality of data pre-processing Quality of atmospheric correction Signal-to-noise ratio estimation
The pre-processing of hyperspectral data includes radiometric correction, georectification and geo-referencing. The owner and operator of the sensor usually takes care of the radiometric calibration before the flight and radiometric correction after the flight. The end-user of hyperspectral data usually has no means to estimate the quality of the radiometric correction. Nevertheless some visual assessment of radiance data can be done in order to find possible anomalies. Quality control of geo-rectification is easier: geometric errors in the data can be found by studying straight lines, for example roads, in the imagery. When a mosaic image is constructed from individual flight lines, possible geometric errors can be seen as discontinuity points in the image. The quality of geo-referencing can be evaluated using ground control points (GCPs). Usually land use authorities have accurate GPCs available for the quality control. If not, other accurate geo-referenced data, e.g. maps or aerial photographs, can be used. The quality of atmospheric correction can be evaluated using REF and PIF targets. Ideally, there should be at least one REF or PIF target for each flight line. The quality of PIF targets should be checked carefully as they are of natural origin; homogeneity of surface material should be checked using adequate number of spectral measurements covering the whole area of the PIF target. There should also be adequate spectral divergence among PIF targets. The spectral measurement of a PIF target should be made simultaneously with the hyperspectral data acquisition to avoid reflectance changes caused by altered weather conditions. PIF targets can also be constructed if adequate number of natural PIFs cannot be found. REF targets should be moved by several field teams in order to have time to cover all flight lines.
Figure 3.4. The estimated SNR of hyperspectral data measured from flight line 07040933 of the Hype08 campaign (SNR is a unitless ratio). The acquisition system can cause different types of degradation leading to loss of image quality. The first type of degradation is radiometric noise caused mainly by photonic effects in the photon detection process by electronic devices and by quantization. This noise can often be assimilated to white noise even if some correlation exists between different bands (Christophe et al. 2005). Other types of degradation are due to the optical characteristics of the imaging spectrometers. The point spread function (PSF) can cause a smoothing effect along the spatial dimension (Liao et al. 2000). The dispersion element and the CCD characteristics can produce a smoothing effect along the spectral dimension. Some post processing, including filtering, compression, or transformations, can also produce some smoothing effects as well as the Gibbs effect. All these types of degradation can be evaluated using signal-to-noise ratio (SNR). SNR is a good generic quality indicator of hyperspectral data. Reduced SNR can indicate sensor malfunctions, poor solar illumination or other acquisition errors. In order to determine the SNR of the Hype08 hyperspectral data, studies were carried out by the Pori Unit of Tampere University of Technology. Several methods for SNR estimation have been proposed in the literature. Green et al. (2003) proposed an image-based method where the most homogenous region in the imagery is identified. The standard deviation of that homogenous region represents the noise characteristics of the hyperspectral data. Roger & Arnold (1996) proposed a method were between-band (spectral) and within-band (spatial) correlations are used to de-correlate the image data via linear regression. The de-correlation leaves noise-like residuals whose variance estimates the noise. The method proposed by Green et al. (2003) was chosen because of its simplicity and clear physical basis, although the method can suffer from the variation of land cover
types in the imagery. The maximum size of a homogenous region was determined by studying the most homogenous region obtained using different sizes. Results showed that there are at least 10*10 m wide homogenous areas in all flight lines. SNR was calculated by dividing the mean of the most homogenous area by its standard deviation. As an example, the estimated SNR of the hyperspectral data measured from flight line 0704-0933 is shown in Figure 3.4. Reduced SNR due to strong atmospheric attenuation in water vapor regions can be seen around wavelengths 940, 1130, 1400 and 1900 nm. SNR is also reduced at the SWIR region around 2400 nm possibly due to poor sensor alignment. Estimated SNR correlates well with sensors specifications indicationg the SNR of 350-500 (peak). The optimal acquisition time on 4th of July was at 10:00 UTC when the solar elevation angle reached the daily maximum. The first flight line started at 06:31 UTC and the last flight line ended at 15:05 UTC. The variation of SNR at different flight lines was minor and there was no correlation between acquisition time and SNR. This indicates that solar irradiation was adequate enough for the acquisition of high quality data despite the notable time offset from the optimum timing. 3.3
Pre-processing of hyperspectral data
The data acquired using an imaging spectrometer contains several errors which are generated due to atmosphere, viewing geometry, platform movements and other sources. These errors can be categorized as radiometric and geometric errors. The effects of both errors are minimized in data pre-processing using several different correction methods. Geometric corrections try to remove geometric distortion whereas radiometric correction transforms recorded values, being just unitless digital numbers, into radiance values corresponding to radiation reflected or emitted from the surface. Nowadays some of these corrections can be automated but most must be performed in the pre-processing of imaged data. Most of the modern imaging spectrometers are push-broom scanners. They record data one line at a time and when aircraft moves forward they record another line thus comprising the total image. The geometric distortion of hyperspectral data is generated by disturbances affecting the platform or the sensor as well as viewing geometry and irregularities of the surface (Lumme 2004). Alterations in aircrafts speed, altitude or direction as well as platform rotations cause distortion in the viewing geometry. When a small single engine aircraft is used, such alterations are frequent. The orbit of the remote sensing satellites is very stable which reduces the need for geometric corrections in space borne data acquisition. The speed, altitude, location and rotations can be registered using a GPS/INS instrument during the data acquisition. This is used in the correction process of geometric distortion. Geometric distortion can be corrected using two different techniques (Richards & Xiuping 1999). The characteristics and the magnitude of error can be modeled thus enabling the removal of the distortion effects in the data. Models for each type of distortion must be created separately. Models can be created if the origin and quality of the distortion is well known. Another technique to correct geometric errors is to co-
register data with ground control points (GCPs) whose location is precisely known. When adequate number of GPCs covering the whole image is used, all geometric errors can be corrected at once even if the origin of some errors is unknown. This technique is widely used and it is easiest to perform. Remote sensing data usually contain noticeable radiometric errors. These errors disturb visual interpretation of imagery and degrade classification results of the data. Radiometric correction involves various steps to model sensor noise, atmospheric conditions and solar illumination effects. Most natural surfaces reflect solar radiation via diffuse reflection, i.e., radiation is reflected to all directions. The intensity of reflected radiation depends on the arrival and departure angles. Natural materials have their own individual reflection properties. The orientation of reflective surface and the sensors positioning in relation to the sun and the target have an effect on the recorded data. Therefore similar targets can get different radiance values depending on their location in the image. This kind of radiometric errors are difficult to correct due to the multitude of contributing factors. The pre-processing of HYPE08 data was done using the GaliGeo software from Specim Ltd. GaliGeo software performs an automated pre-processing where raw AISA sensor data are transformed into a radiometrically corrected, georeferenced image. In addition to raw data, GaliGeo needs the data recorded by the GPS/INS and the FODIS instruments. The FODIS instrument records the onboard solar irradiance during the acquisition. The preliminary results of pre-processing were not satisfactory as there were remarkable geometric errors visible in the data. The University of Debrechen, Hungary, consulted the issue with Specim Ltd. and they found out that it was necessary to use new boresight parameters. The boresight parameters compensate the misalignment between the sensor and GPS/INS instrument. The pre-processing results using the new parameters were good, i.e., without any noticeable errors.
Atmospheric correction of hyperspectral data
The objective of atmospheric correction is to retrieve the surface reflectance (that characterizes the surface properties) from remotely sensed imagery by removing the atmospheric effects. After radiometric correction the measured values are expressed in radiance, i.e., the energy reflected from a surface area (W/m²). In atmospheric correction, the radiance values are converted into reflectance data, measuring the fraction of radiation reflected from the surface. Atmospheric correction is a difficult procedure due to the complex nature of the atmosphere; the correction procedure must be done individually for each flight line. Atmospheric correction should be done with utmost care because it largely determines the usability of the final data. The application of most algorithms and indices requires well calibrated reflectance data. Accurate change detection cannot be accomplished without atmospheric correction because otherwise it is impossible to determine whether the change occurred in the continuously varying atmosphere conditions or in the target under study. In Figure 3.5 atmospheric features in radiance spectra without atmospheric correction are shown: the water vapor absorption features are at approximately 940 and
1135 nm, and the oxygen and the carbon dioxide absorption features at 760 nm and 2000 nm, respectively. Opaque atmospheric regions around 1400 and 1900 nm, where virtually no radiation is passed through the atmosphere, are clearly visible. Several methods for the atmospheric correction have been proposed in the literature. They can be divided into two categories: (1) empirical and (2) model-based methods. The empirical methods rely on the scene information and do not use any physical model as the model-based methods do. The scene information means the information that is embedded in the image, i.e., the radiance at certain location. There are empirical-based methods that rely on the raw scene data without ground reference information whereas some methods rely on the raw scene data together with ground reference information. There are two common approaches that do not use the ground reference information, the Internal Apparent Relative Reflectance (IARR) method (Kruse et al. 1985) and the Flat Field (FF) method (Roberts et al. 1986). In both methods, the spectral data of each pixel is divided by a reference spectrum. In the FF method the reference spectrum is from a homogenous bright target and in the IARR method an average scene spectrum is used as the reference spectrum. The drawback of these methods is that they are prone to artifacts and strongly dependent on the landscape of the target (Ben-Dor et al. 2004).
Figure 3.5. Atmospheric features in radiance spectra.
Most recentaddition to empirical methods that do not use ground reference information is the Quick Atmospheric Correction (QUAC) method. The QUAC is based on the empirical finding that the average reflectance of a collection of diverse material spectra, such as the endmember spectra in a scene, is essentially scene-independent (Bernstain et al. 2005). The use of QUAC has some restrictions, e.g., there must be a certain minimum amount of land area in the scene. The most widely used empirical method is the Empirical Line (EL) approach (Conel et al. 1987). The EL approach requires field measurements of reflectance spectra for at least one bright target and one dark target. The imaging spectrometer data over the surface targets are linearly regressed against the field-measured reflectance spectra to derive the gain and offset curves. The curves are then applied to the whole image for the derivation of surface reflectances for the entire scene. This method produces spectra that are most comparable to reflectance spectra measured in the field. However, if changes occur in the atmospheric properties outside the area for which ground data is available, which is often the case, the spectral reflectance data will contain atmospheric features. Model-based correction approaches use methods in which the radiance at the sensor is modeled using radiative transfer models and the data from detailed atmospheric and sun information archives (e.g., MODTRAN, HITRAN2000). In this procedure, field measurements are not required and only the basic information such as the site location and elevation, flight altitude, sensor model, local visibility and acquisition times, are required. Several model-based methods dedicated to retrieving reflectance information from hyperspectral data have been developed, such as ATREM (Gao et al. 1993), ATCOR (Richter 1996), CAM5S (O’Neil et al. 1996), FLAASH (Adler-Golden et al. 1998), HATCH (Qu et al. 2000) and ACORN (Kruse 2004). All the methods are quite similar in their basics and operation. ATCOR, FLAASH and ACORN are based on the use of MODTRAN 4 radiative transfer code. In order to determine the most accurate atmospheric correction method to be used in the processing of the HYPE08 data, a study was conducted at Tampere University of Technology Pori unit: the measured reflectance spectra of REF and PIF targets were used as references for the comparison of corrected data. REF targets are spectrally homogenous tarpaulins laid to the ground during the overflight. Pseudo Invariant Feature (PIF) targets are natural homogenous areas on the ground surface. Potential PIF targets were searched in the planning phase of HYPE08 using satellite- and aerial photographs available. PIF sites were selected using 3 different criteria: surface material should be homogenous, the size of the area should be at least 10*10 m² and the material should be robust against moisture variations. An example of a PIF site is shown in Figure 3.6. The reflectance of the PIF target was measured at the same time as the overflight took place. REF targets were white and black tarpaulins of 15*15 m² which were moved from one flight line to another. The material used in REF targets is carefully selected in order to ensure smooth and stable spectral properties. Black and white tarpaulins used as reference targets are shown in Figure 3.7.
Figure 3.6. Measurement of PIF 028 site along flight line 0704-0827 of the HYPE08 campaign using portable field spectrometer. PIF 028 is a homogenous gravel road near Olkiluoto. Photo: Ari Ikonen.
Figure 3.7 Black and white tarpaulins used as reference targets were moved from one flight line to another during the HYPE08 flight campaign. The location is at the Luvia fishing harbour. Photo: Arto Vuorela.
Two empirical methods (EL, QUAC) and two model-based methods (FLAASH, ATCOR) were tested. Corrected reflectance data were compared against two independent REF and three independent PIF targets. The EL and ATCOR methods produced promising results while the other methods failed. Unfortunately there were not enough PIF and REF targets for the use of EL correction over the entire acquisition area. The use of EL correction ideally requires at least one bright and one dark REF or PIF spectrum per flight line. One bright and one dark reference per two flight lines could yet be considered to be adequate number of references, but this demand could not be satisfied over the whole flight area. The collection of adequate number of REF and PIF spectra is almost impossible in flight lines covering the coastal water area and it is not safe to assume that it could be done during future flight campaigns either. Considering the restrictions on the use of the EL method, the ATCOR method was chosen for the atmospheric correction. The use of the ATCOR requires the definition of several physical parameters related to data acquisition. Some physical parameters such as flight altitude are easily available, but some parameters were not available or were unclear. Tests were conducted in order to determine which parameters should be used for atmospheric type, water vapor content and visibility. Corrected reflectances were simply compared against known references using different values for the parameters systematically. The use of some ATCOR options was tested as well. The atmospheric correction was done using the following parameters: -
Atmospheric type: Rural Water vapor column: 2.0 cm Adjacency range: 0.2 km Visibility: 45 km Water vapor algorithm: No water vapor correction. All algorithms available failed because of noisy water vapor regions in the data. Variable visibility: Yes Variable water vapor: Yes Shadow Removal: No
The average RMSE error of the ATCOR-corrected reflectance data was 6.8 % when compared to 5 reference spectra, i.e., PIF and REF targets. The corrected spectrum of a white reference tarpaulin is shown in Figure 3.8. The most distinct difference between corrected and measured spectra is in the wavelengths from 1000 nm to 1100nm. This is most likely due to neighboring water vapor regions. A corrected spectrum of a natural PIF target (red synthetic pavement of a sports stadium) is shown in Figure 3.9. The accuracy of atmospheric correction can be considered very satisfactory when compared to those published in the literature (Ben-Dor et al. 2004).
Figure 3.8. The corrected spectra (white) and measured spectra (green) of white reference tarpaulin on flight line 0704-0908. The blue areas are water vapor regions removed by the ATCOR software.
Figure 3.9. The corrected (white) and measured (green) spectra of red synthetic pavement of a sports stadium on flight line 0704-1024 of HYPE08. The blue areas are water vapor regions removed by the ATCOR software.
Change detection is the process of identifying differences in the state of an object or phenomenon by observing it at different times. Essentially, it involves the ability to quantify temporal effects using multi-temporal data sets. Change detection is useful in such diverse applications as land use change analysis, monitoring of shifting cultivation, assessment of deforestation, study of changes in vegetation phenology, seasonal changes in pasture production, damage assessment, crop stress detection, disaster monitoring, snow-melt measurements, day/night analysis of thermal characteristics and other environmental changes (Singh 1989). Manual handling of data for change detection using sequential imagery is a huge task. Especially in the case of hyperspectral data it is practically impossible due to the amount of the data. Many change detection techniques have been developed. Previous literature has shown that image differencing, principal component analysis and post-classification comparison are the most common methods used for change detection. In recent years, spectral mixture analysis, artificial neural networks and integration of geographical information system and remote sensing data have become important techniques for change detection applications (Lu et al. 2003). Different change detection algorithms have their own merits and no single approach is optimal and applicable to all cases. In practice, different algorithms are often compared to find the best change detection results for a specific application. Singh (1989) listed the following change detection algorithms in his comprehensive review article: -
Image Thresholding Image Differencing Image Rationing Image Regression Change Vector Analysis Vegetation Index Differencing Multi-date Principal Component Analysis Post-classification Technique
Thresholding of an image is used to determine whether the change is significant or not. Thresholding is included in many change detection methods. Pixels whose value exceeds determined level are simply highlighted in the image. Several different thresholds can be defined in order to provide information on the magnitude of the change. In image differencing method, two spatially registered images are subtracted pixel by pixel on each band to produce a new change image between two points in time. In the image rationing method two co-registered images are rationed pixel by pixel on each band as in the previous method. The unchanged area is characterized by a ratio values close to 1. Depending on the nature of changes between two points of time, the changed areas have higher or lower values (Singh 1989). The image regression method is based on the assumption that pixels from a time point are in a linear function of the pixels of another time point. Therefore change detection can be done using linear regression and the regression line can be determined by the least squares method.
Deviations from the regression line are interpreted as changes. Image regression has produced slightly better results than image differencing in detecting urban land cover changes and tropical forest cover changes (Singh 1989). The change vector analysis considers the magnitude and the direction of the spectral change vector (Lu et al. 2003). When vegetation undergoes a change its spectral properties change accordingly. The vector describing the direction and magnitude of change from the first to the second date is a spectral change vector. The decision that a change has occurred is made if the magnitude of the computed spectral change vector exceeds a specified threshold criterion. The direction of the vector contains information about the type of change, i.e. the decrease or increase of vegetative vigor. Vegetation Indices (VIs) have long been used in remote sensing for monitoring temporal changes associated with vegetation (Tuominen et al. 2009). Some VIs such as the Normalized Difference Vegetation Index (NDVI) measure the general quantity and vigor of vegetation whereas some VIs such as the Carotenoid Reflectance Index (CRI) measuring the carotenoid pigment content are very specific. Differencing the VIs calculated from data recorded at different times allows the study of specific vegetative changes when an appropriate VI is used. The principal components analysis technique is used to reduce the number of spectral components to fewer principal components accounting for the most of the variance in the original multispectral images. In multi-temporal studies the principal components for two or more dates are often compared as in image differencing method. Alternatively, two images of the same area, which are recorded at different dates, can be superimposed and treated as a single image where the number of the bands equals the sum of the number of bands of original images (Singh 1989). The principal component analysis of this superimposed data set should result in the gross differences associated with overall radiation and albedo changes in the major component images and statistically minor changes associated with local changes in land surface appearing in the minor component images. In the post-classification approach, images belonging to different dates are classified and labeled individually. Later, the classification results are compared directly and the areas of changes are detected. Supervised and unsupervised classification methods are used in this approach. Individual classification of two images at different dates minimizes the problem of normalizing for atmospheric and sensor differences between the data of the two dates. In addition to the change detection methods presented above, more advanced methods such as hybrid change detection, artificial neural networks, Li-Strahler reflectance model, spectral mixture model and biophysical parameter model have been presented in the literature (Lu et al. 2003). The use of these methods offers great possibilities in change detection, although there are some restrictions which might limit their utilization. In general, advanced methods are complicated and computationally demanding. Processing of high volume hyperspectral data may become unfeasible, e.g., the size of the HYPE08 data set is over 500 gigabytes. Some advanced methods require large amount of training data or field measurements that are hard to accomplish.
PROPOSAL FOR HYPERSPECTRAL MONITORING OF OLKILUOTO REPOSITORY SITE
In this chapter a proposal for a hyperspectral monitoring system for the Olkiluoto repository site is described. The system is based on consecutive hyperspectral airborne campaigns. The data acquisition is followed by precisely defined and well documented data processing producing accurate and reliable information of the environment around the Olkiluoto site.
Hyperspectral monitoring system
The system is based on airborne hyperspectral remote sensing. The baseline data used as a reference is the data recorded during HYPE08 campaign. Airborne campaigns are continued at regular intervals, e.g., every third or fourth year to update the hyperspectral data sets. In the years between airborne campaigns hyperspectral data is acquired from other sources, e.g., satellite borne data or by using UAV sensors in order to reduce the data acquisition costs. A comprehensive field campaign is carried out during the airborne flight campaign in which reference data is collected. The optimal acquisition time is in July when the greenness of vegetation is full blown and the statistical possibility of good weather conditions is highest. The work flow of data processing in hyperspectral monitoring system is presented in Figure 4.1. It is important that all methods used in the data processing are well defined and documented. The same methods are applied to both data sets used in change detection. This ensures that detected changes are true changes in environment and not the changes due altered methods. It is possible to update processing methods when the original raw data is stored from each data acquisition. The quality of data acquisition, pre-processing and atmospheric correction is controlled. Data is not used in monitoring system if the quality standards are not met. The change detection is done using several different methods. Different types of environmental changes are detected using methods most suitable for the case. Hyperspectral imaging enables to monitor several key observables representing environmental state. Automatic change detection can produce information on the location, type and magnitude of the change. The automated change detection is followed by expert interpretation of the change. The expert interpretation defines the biological meaning of the detected change. Field verification and measurements are used in the expert interpretation process. More research is needed in order to further develop the change detection process. The amount of expert interpretation can be reduced by developing the automated methods. Vegetation mapping is done using a spectral library containing reflectance spectra collected in field measurements. Vegetation maps contain information on vegetation quality and quantity.
Figure 4.1. The work flow of data processing in the proposed hyperspectral monitoring system.
Integration with other environmental data
During recent years Posiva has performed comprehensive environmental studies in and around the Olkiluoto site. These studies include: -
Results of these studies can be integrated with results from the hyperspectral monitoring system. Results from the hyperspectral monitoring system can be validated using other environmental data. Results can also be used as training data for hyperspectral classification methods. Integration can be used both ways. Hyperspectral data can be used to assess if the spatial coverage of studies based on sampling is adequate. Most of the environmental studies are published as working reports and synthesized in the Olkiluoto Biosphere Description reports (e.g. Haapanen et al. 2009). A map showing the main vegetation types on Olkiluoto Island is presented in Figure 4.2 as an example of environmental data that can be used with the hyperspectral monitoring system. The variation within the vegetation classes can be studied also with help of the hyperspectral data.
Figure 4.2. The main vegetation classes on Olkiluoto Island (based on the data of Miettinen & Haapanen 2002).
In this report a review on hyperspectral remote sensing techniques and applications have been presented. A brief introduction on general remote sensing principles is followed by detailed review on hyperspectral applications. A comprehensive overview on remote sensing applications for nuclear safeguards is also presented. In chapter 3 various phases in hyperspectral data processing have been explored. The chapter also includes a detailed description of HYPE08 flight campaign. Finally a proposal for hyperspectral monitoring system for the Olkiluoto repository site is presented. Based on the review presented in this report it can be concluded that hyperspectral remote sensing is very promising technique in the monitoring of nuclear fuel repository site.
REFERENCES Absalom, J. P., Young, S.D., Crout, N.M., Sanchez, A., Wright, S.M., Smolders, E., Nisbet, A.F. & Gillett, A.G. 2001. Predicting the transfer of radiocaesium from organic soils to plants using soil characteristics. Journal of Environmental Radioactivity, 52: 31– 43. Asner, G. P. (2008). Hyperspectral Remote Sensing of Canopy Chemistry, Physiology, and Biodiversity in Tropical Rainforests, In: Hyperspectral Remote Sensing of Tropical and Sub-Tropical Forests, Kalacska, M. & Sanchez-Azofeifa, G.A., (1.), 261-296, CRC Press, ISBN 9781420053418, Boca Raton Asrar, G. 1989. Theory and applications of optical remote sensing. New York, John Wiley and sons Ltd. 734 p. Barret, E. C. & Curtis, L. F. 1992. Introduction to environmental remote sensing. London, Chapman & Hall Ltd. 413 p. Ben-Dor, E. & Banin, A. 1994. Visible and near infrared (0.4-1.1 um) analysis of arid and semiarid soils. Remote Sensing of Environment, 48:261-274 Ben-Dor, E., Kindel, B. & Goetz, A.F.H. 2004. Quality assessment of several methods to recover surface reflectance using synthetic imaging spectroscopy data. Remote Sensing of Environment, 90:389-404. Bernstein, L. S., Adler-Golden, S.M., Sundberg, R.L., Levine, R.Y., Perkins, T.Y, Berk, A., Ratkowski, A.J., Felde, G. & Hoke, M.L. 2005. Validation of the QUick atmospheric Correction (QUAC) algorithm for VNIR-SWIR multi- and hyperspectral imagery. SPIE Proceedings, Algorithms and Technologies for Multispectral, Hyperspectral,and Ultraspectral Imagery XI. Vol. 5806. Borstad, G.A., Truong, Q.S.B., Keeffe, R., Staenz, K. & Neville, R. 2001. Using limnological and optical knowledge to detect discharges from nuclear facilities – potential application for international safeguards. Presentation at the IAEA symposium on international safeguards, Vienna, Austria Borstad, G.A., Lim, J., Brown, L.N., Truong, Q.S.B., Healey, G., Kelly, R., Staenz, K. & Neville, R. 2006 Satellite hyperspectral imaging in support of nuclear safegaurds monitoring. Proceedings of 47th INMM Annual Meeting, Nashville, Tennesee, 16-20 July 2006 Borstad, G.A., Lim, J., Brown, L.N., Truong, Q.S.B., Healey, G., Kelly, R., Staenz, K. & Neville, R. 2007 Satellite hyperspectral imaging in support of nuclear safegaurds applications. Proceedings of 48th INMM Annual Meeting, Tucson, Arizona, 8-12 July 2007
Bowman, A.P., Winter, E.M., Stocker, A.D. & Lucey, P.G. 1998. Hyperspectral infrared techniques for buried landmine detection. Proceedings of the second international conference on the detection of abandoned land mines, Edinburg, UK, 1214 Oct. Burger, J. 2007. A model for selecting bioindicators to monitor radionuclide concentrations using Amchita Island in the Aleutians as a case study. Environmental Research, 105:316-323 Cavalli, R.M., Colosi, F., Palombo, A., Pignatti, S. & Poscolieri, M. 2007. Remote hyperspectral imagery as a support to archaeological prospection. Journal of Cultural Heritage, 8:272-283 Chabrillat, S., Goetz, A. F. H., Krosley, L., & Olson, H. W. 2002. Use of hyperspectral images in the identi fi cation and mapping of expansive clay soils and the role of spatial resolution. Remote Sensing of Environment, 82: 431445. Chen, C., Shi, P. & Mao, Q. 2003. Application of remote sensing techniques for monitoring the thermal pollution of cooling-water discharge from nuclear power plant. Journal of environmental science and health, 38:16591668. Christophe, E., Leger, D. & Mailhes, C. 2005. Quality Criteria Benchmark for hyperspectral imagery. IEEE transactions on geosciences and remote sensing, 43:21032114 Conel, J. E., Green, R. O., Vane, G., Bruegge, C. J., Alley, R. E., & Curtiss, B. J. 1987. Airborne imaging spectrometer-2: Radiometric spectral characteristics and comparison of ways to compensate for the atmosphere. Proceedings of SPIE, 834: 140– 157. Curran, P.J. 1989. Remote sensing of foliar chemistry. Remote Sensing of Environment, 30:271-278 Davids, C. & Tyler, A.N. (2003). Detecting contamination-induced tree stress within the Chernobyl exclusion zone. Remote Sensing of Environment, 85:30-38. Dekker, A. 1993. Detection of the optical water quality parameters for eutrophic waters by high resolution remote sensing. Doctoral thesis . Free University, Amsterdam Ellis, J., Davis, H.H., Zamudio, J.A. 2001. Exploring for onshore oil seeps with hyperspectral imaging. Oil Gas Journal. 37:49–55 Farley, V., Vallieres, A., Chamberland, M. & Villemaire, A. 2006. Performance of the FIRST, a longwave infrared hyperspectral imaging sensor. Proceedings of SPIE optically based biological and chemical detection for defence , Stockholm, Sweden, 1113 Sep. Gao, B. -C., Heidebrecht, K. B., & Goetz, A. F. H. 1993. Derivation of scaled surface reflectance from AVIRIS data. Remote Sensing of Environment, 44:145–163.
Gittins, G.M. & Marinelli, W.J. 1998. LWIR Multispectral Imaging Chemical Sensor. Proceedings of SPIE air monitoring and detection of chemical and biological agents symposium, Boston, MA, 1-5 Nov. Goetz, A. F. H., & Srivastava, V. 1985. Mineralogical mapping in the Cuprite mining district, Nevada. Proceedings of the Airborne Imaging Spectrometer Data Analysis Workshop. Pasadena, CA: JPL Publication 85—41, Jet Propulsion Laboratory. Goetz, A.F.H. 2009. Three decades of hyperspectral remote sensing of the Earth: A personal view. Remote Sensing of Environment, 113:s5-s16 Gordo, O. & Sanz, J.J. 2005. Phenology and climate change: a long-term study in a Mediterranean locality. Oecologia, 146: 484–495. Green, R.O., Pavri, B.E. & Chrien, T.G. 2003. On-orbit radiometric and spectral calibration characteristics of EO-1 Hyperion derived with an underflight of AVIRIS and in situ measurements at Salar de Arizaro, Argentina. IEEE Transactions on geosciences and remote sensing, 41: 1194–1203. Guanter, L., Segl, K. & Kaufmann, H.. 2009. Simulation of optical remote sensing scenes with application to the EnMAP hyperspectral mission. IEEE Transactions on geosciences and remote sensing, 47:2340-2351 Haapanen, R., Aro, L., Helin, J., Hjerpe, T., Ikonen, A.T.K., Kirkkala, T., Koivunen, S., Lahdenperä, A.-M., Puhakka, L., Rinne, M. & Salo, T. 2009. Olkiluoto Biosphere Description 2009. Posiva Oy, POSIVA 2009-02. Hjerpe, T., Ikonen, A.T.K. & Broed, R. 2010. Biosphere assessment report 2009. Posiva Oy, POSIVA 2010-03. Holma, H., Mattila, A.-J. & Hyvärinen, T. 2009. New thermal infrared hyperspectral imagers. NATO Technical report RTO-SET-151, 2009 Häme, T. 2003. Use of satellite and airborne remote sensing in the safeguards of a nuclear waste repository site. In: Okko, O. (ed.). Safeguards for final disposal of spent nuclear fuel. Methods and technologies for the Olkiluoto site. STUK-YTO-TR 199. Helsinki: Radiation and Nuclear Safety Authority. Ilus, E. 2009. Environmental effects of thermal and radioactive discharges from nuclear power plants in the boreal brackish-water conditions of the northern Baltic Sea. STUKA238. Helsinki: Radiation and Nuclear Safety Authority. Kirkland, L. E., Herr, K.C., Keim, E.R., Adams, P.M. Salisbury, J.W., Hackwell, J.A. & Treiman, A. 2002. First Use of an Airborne Thermal Infrared Hyperspectral Scanner for Compositional Mapping, Remote Sensing of Environment, 80: 447–459.
Kokaly, R.F., Despain, D.G., Clark, R. N. & Livo, K.E. 2003. Mapping vegetation in Yellowstone National Park using spectral feature analysis of AVIRIS data. Remote Sensing of Environment, 84:437-456 Kruse, F. A., Raines, G. I., & Watson, K. 1985. Analytical techniques for extracting geologic information from multichannel airborne spectroradiometer and airborne imaging spectrometer data. Proceedings of the 4th thematic conference on remote sensing for exploration geology. Ann Arbor, MI. Kruse, F.A, Lefkoff, A.B. & Dietz, J.B. 1993. Expert system-based mineral mapping in northern death valley, California/Nevada, using the Airborne Visible/Infrared Imaging Spectrometer (AVIRIS). Remote Sensing of Environment, 44:309-336 Kruse, F. A. 2004. Comparison of ATREM, ACORN, and FLAASH atmospheric corrections using low-altitude AVIRIS data of Boulder, CO. Summaries of 13th JPL Airborne Geoscience Workshop, Jet Propulsion Lab, Pasadena, CA. Levesque, J. Neville, R.A., Staenz, K. & Truong, Q.S. 2001. Preliminary results on the investigation of hyperspectral remote sensing for the identification of uranium mine tailings. In: Proceedings of the ISSSR: June 10-15, 2001, Quebec City, Canada Liao,L.B., Jarecke, P.J., Gleichauf, D.A. & Hedman, T.R. 2000. Performance haracterization of the Hyperion imaging spectrometer instrument. Proceedings of SPIE Conference on Earth Observing Systems V, vol. 4135, W. L. Barnes, Ed., Nov. 2000. Lichtenthaler, H. K. 1996. Vegetation stress: an introduction to the stress concepts in plants. Journal of Plant Physiology, 148: 4– 14. Lindfors, A., Huttunen, O. & Mykkänen, J. 2008. Vedenlaadun alueellinen jakauma Olkiluodon edustalla heinäkuussa 2008. Eurajoki 2008, Posiva Oy, Working report 2008-86. 24 s. + 6 s. appendices. Lu, D., Mausel, P., Brondiacutezio, E. & Moran, E. 2003. Change Detection Techniques. International Journal of Remote Sensing, 25:2365-2407. Lucey, P.G., Williams, T.J., Winter, E. & Winter, M. 2001. Two years of operations of AHI: a Lwir hyperspectral imager. Proceedings of IGARSS 2001 International Geoscience and Remote Sensing Symposium, Sydney, Australia. Lumme, J. 2004. Vegetation and soil classification and analysis from imaging spectrometer data. Master thesis. Helsinki university of technology, Helsinki da Luz, B.R. & Crowley, J.K. 2010. Identification of plant species by using high spatial and spectral resolution thermal infrared (8.0-15.5 um) imagery. Remote Sensing of Environment, 114:404-413
MacDonald, J. S., Ustin, S. L. & Schaepman, M. E. 2009. The contributions of Dr.Alexander F.H. Goetz to imaging spectrometry. Remote Sensing of Environment, 113:s2-s4 Mares, A.G., Olsen, R.C. & Lucey, P.G. 2004. LWIR Spectral measurements of volcanic sulfur dioxide plumes, Proceedings of the SPIE, Volume 5425, pp. 266-272. Mars, J.C. & Crowley, J.K. 2003. Mapping mine wastes and analyzing areas affected by selenium-rich water runoff in southeast Idaho using AVIRIS imagery and digital elevation data. Remote Sensing of Environment, 84:422-436 Miettinen, N. & Haapanen, R. 2002. Vegetation types on Olkiluoto Island. Posiva Oy, Working Report 2002-54. 54 p. Monteiro, S.T, Minekawa, Y., Kosugi, Y & Kunio, O. 2007. Prediction of sweetness and amino acid content in soybean crops from hyperspectral imagery. ISPRS Journal of Photogrammetry and Remote Sensing, 62:2-12 Moore, B. & Allard, G. 2008. Climate change impacts on forest health, Working paper FBS/34E02/8, 38p. FAO, Rome. Neville, R.A., Staenz, K., Levesque, J., Nadeau, C., Truong, Q.S. & Borstad, G.A. 2001. Uranium mine detection using an airborne imaging spectrometer. In: Fifth International Airborne Remote Sensing Conference: September 17-20, 2001; San Francisco, CA Niemeyer, I. 2009. Perspectives of satellite imagery analysis for verifying the nuclear non-proliferation treaty, In : International safeguards and satellite imagery, Jasani, B., Niemeyer, I., Nussbaum, S., Richter, B. & Stein, G., Springer, ISBN 9783540791317, Berlinjk Nolin, A. W. & Dozier, J. 1993. Estimating snow grain-size using AVIRIS data. Remote Sensing of Environment, 44, 231238. O’Neil, N.T., Royer, A., & Nguyyen, M.N. 1996. Canadian Advanced Modified 5S (CAM5S). Internal Report, CARTEL-1996-0202, Universitiede Sherbrooke, Sherbrooke, Quebec, Canada. 46 pp. Pabich, P. J. 2002. Hyperspectral imagery: warfighting through a different set of eyes. Maxwell Alabama, Air University USAF. 59 pp. Papastefanou, C., Manolopoulou, M., Stoulos, S., Ioannidou, A., & Gerasopoulos, E. 1999. Soil-to-plant transfer of 137Cs, 40K and 7Be. Journal of Environmental Radioactivity, 45:59– 65. Penna, B., Tillo, T., Magli, E. & Olmo, G. 2001. Quality assesment for hyperspectral imagery: comparison between lossy and near-lossless compression. Prodeedings of
Asilomar conference on signals, systems and computers. Oct.29-Nov.1, 2006; Pacific Grove, CA Peterson, C. H., Rice, S. D., Short, J. W., Esler, D., Bodkin, J. L., Ballachey, B. E. & Irons, D. B. 2003. Long-term ecosystem response to the Exxon Valdez Oil Spill. Science, 302:2082-2086 Puckrin, E. & Theriault, J-M. 2004. Passive standoff detection of radiological products by Fourier-transform infrared radiometry. Optics Letters, 29:1375-1377 Puckrin, E., Theriault, J-M., Lavoic, H. & Levesque, J. 2006. Recent advances in the remote sensing of radiological materials by passive FTIR radiometry, DRDC-Valcartier, Valcartier, PQ, Canada, March 2006 Qu, Z., Goetz, A. F. H., & Heidbrecht, K. B. 2000. High accuracy atmosphere correction for hyperspectral data (HATCH). Proceedings of the 9th JPL airborne earth science workshop. JPL-Pub., vol. 00-18 (pp. 373– 381). Pasadena, CA Richards J. A. & Xiuping J. 1999. Remote sensing digital image analysis: an introduction. 3rd, revised and enlarged edition. Berlin, Springer. 363 s. ISBN 3-54064860-7 Richter, R. 1996. A spatially adaptive fast atmosphere correction algorithm. International Journal of Remote Sensing, 11: 159– 166. Richter, R. 2005. Hyperspectral sensors for military applications. NASA technical report RTO-MP-SET-094, 2005. Roberts, D. A., Yamaguchi, Y., & Lyon, R. J. P. 1986. Comparison of various techniques for calibration of AIS data. Proceedings of the 2nd AIS workshop. JPL Publication, vol. 86-35. Pasadena, CA. Roger, R.E. & Arnold, J.F. 1996. Reliably estimating the noise in AVIRIS hyperspectral images. International Journal of Remote Sensing, 17: 1951-1962. Saari, H., Aallos, V., Akujärvi, A., Antila, T. Holmlund, C., Kantojärvi, U., Mäkynen, J. & Ollila, J. 2009. Novel Miniaturized Hyperspectral Sensor for UAV and Space Applications, in Proceeding of SPIE Europe Remote Sensing 2009 Conference , Berlin, Germany Salem, F & Kafatos, M.2001. Hyperspectral image analysis for oil spill mitigation. Proceedings of the 22nd Asian Conference on Remote Sensing, 5-9 November 2001, Singapore Schaepman, M. E:, Ustin, S. L., Plaza, A. J., Painter, T.H., Verrelst, J. & Liang, S. 2009. Earth system science related imaging spectroscopy – An assessment. Remote Sensing of Environment,113:S123-S137
Siegal, B. S., & Goetz, A. F. H. 1977. Effect of vegetation on rock and soil type discrimination, Photogrammetry Engineering and Remote Sensing, 43, 191196. Singh, A. 1989. Digital change detection techniques using remotely sensed data. International Journal on Remote Sensing, 10:989-1003.. Smartt, H.A., Blair, D., Stork, C.L., Smith, J. & Jasani, B. 2005. Evaluation of remote sensor systems for monitoring uranium mines. Proceedings of the ESARDA European Safeguards Research and Development Association 27th Annual Meeting, London, UK, 10-12 May. Swayze, G. A., Smith, K. S., Clark, R. N., Sulley, S. J., Pearson, R. M. & Vance, J. S. 2000. Using imaging spectrometry to map acidic mine waste. Environmental Science and Technology, 34, 4754. Truong, Q.S., Borstad, G.A., Staenz, K., Neville, R., Leslie, R., Riggs, P. & Bragin, V. 2003. Multispectral and Hyperspectral Imagery for Safeguards and Verification of Remote Uranium Mines. In: Proceedings of the 44th Annual Meeting of the Institute of Nuclear Materials Management (INMM), Phoenix, Arizona. Tuominen, J., Lipping, T. ; Kuosmanen, V : & Haapanen, R. 2009. Remote Sensing of Forest Health, In: Geoscience and Remote Sensing, Pei-Gee Peter Ho, (1.), 29-52, InTeh, ISBN 9789533070032, Vukovar Williams, D.J., Rybicki, N.B., Lombana, A.V., O’Brien, T.M. & Gomez, R.B. 2003. Preliminary investigation of submerged aquatic vegetation mapping using hyperspectral remote sensing. Environmental Monitoring and Assessment, 81:383-392 Vaughan, R.G., Calvin, W.M. & Tranik, J.V. 2003. SEBASS hyperspectral thermal infrared data: surface emissivity measurement and mineral mapping. Remote Sensing of Environment, 85:48-63 Vitousek, P. M., Mooney, H. A., Lubcheno, J. & Melillo, J. M. 1997. Human Domination of Earth’s Ecosystems. Science, 277:494-499