Download to read offline
![R. Al-Tahir et al.: Advancing the Use of Earth Observation Systems 10
data, and image-processing and classification
approaches used (Lu and Weng, 2007).
4.2 Extraction of Information from High
Resolution Images
Within recent years, there has been increased
availability and wide use of high resolution imagery
in land applications. High resolution imagery shows
object information such as structure, texture and
detail clearly, making it ideal for observing object
detail changes on the earth surface and for
monitoring the extent to which humans have altered
the environment (Su and Hu, 2004).
It has been suggested that the application of
traditional per pixel classification methods has
limited applicability to high spatial resolution data
because they cannot fully exploit its content (Hester
et al., 2008). This is especially problematic in
heterogeneous environments where pixel values are
near similar and is in part due to the limited number
of spectral information in such imagery (Cots-Folch
et al., 2007). Additionally, high resolution imagery
can be affected to a great deal by artefacts, such as
shadows, making previous approaches to image
classification unsuitable for this type of imagery.
Image complexity and large data volumes are other
general issues associated with the use of high
resolution imagery that have been reported (Hester et
al., 2008).
One way of increasing classification accuracy
with high resolution imagery is by using approaches
that utilise the texture in the image. Texture is a
repeated variation of intensity and colour that is
directly portraying object structure and space
arrangement in the image (Su and Hu, 2004).
Research incorporating the use of texture measures
to improve spectral classification accuracy of land
cover has already met positive results (Palubinskas et
al., 1995; Franklin et al., 2000; Puissant et al., 2005).
A review of some of these approaches has been
highlighted in (Cots-Folch et al., 2007). Those
strategies targeted towards high resolution imagery
include examples presented in De Martino et al.
(2004) using a partial classification method in the
detection of objects in an urban part of Brazil using
4m Ikonos, and Ettarid et al. (2008) in which an
automated method for extracting building from 2.5m
Spot and Quickbird imagery for the cities Benir
Amir and Rabat in Morocco was used.
Several methods use texture for image
segmentation and classification; they differ mainly
by the degree of prior information they require and
the way texture measures are applied. Commonly
used methods use statistical approaches. These are
based on the measurement of the occurrences of each
grey level value in a particular neighbourhood (Grey
Level Co-occurrence Matrix) (Cots-Folch et al.,
2007). Haralick et al. (1973) suggested a set of
several features, which can be used to classify
texture images; angular second moment, contrast,
correlation, sum of squares, inverse difference
moment, sum average, sum variance, sum entropy,
entropy, difference of variance, difference of
entropy, information measure of correlation 1 and
information measure of correlation 2. Subgroups of
these features have been widely used in research for
a wide array of image classification studies.
Other texture-based methods are embedded in
other schema such as artificial neural networks
(ANNs) and fuzzy classifiers (Shah and Gandhi,
2004; Cots-Folch et al., 2007). The ANN structure is
based on the human brain’’s biological neural
processes. Interrelationships of variables that are
correlated in the image symbolically represent the
interconnected processing of neurons of the human
brain used to develop models. With fuzzy
classification, there are no hard boundaries dividing
geographic objects. Fuzzy classification methods
assign gradual membership of pixels to classes as
degrees in [0, 1], giving the flexibility to represent
pixels that belong to more than one class. A review
of some these applications can be found in Smits and
Annoni (1999).
5. Developing the Methodology
In its approach to assess the sustainability of the
development in Trinidad and Tobago, this study has
developed a methodology to quantify and analyse the
interaction between natural and urban development
in Trinidad and Tobago. As outlined in Figure 1, the
methodology’’s main thrust is the use of a series of
satellite images covering the period from the 1970’’s
to present. These images were acquired and analysed
to depict the nature of land use/cover during
different times.
Land cover information would be extracted from
these images and combined with other available
environmental, demographic, and economical data in
order to define a set of mainly spatially-based
sustainability indicators. A specific set of indicators;
namely land use and settlement patterns, vegetation
cover, loss cover, and fragmentation of land and
habitat are purposely chosen because they can be
extracted and updated, directly or indirectly, using](https://image.slidesharecdn.com/al-tahiretal2009-171207073135/75/Advancing-the-Use-of-Earth-Observation-Systems-for-the-Assessment-of-Sustainable-Development-5-2048.jpg)
More Related Content
Similar to Advancing the Use of Earth Observation Systems for the Assessment of Sustainable Development
Advancing the Use of Earth Observation Systems for the Assessment of Sustainable Development
- 1. R. Al-Tahir etal.: Advancing the Use of Earth Observation Systems 6 ISSN 1000 7924 The Journal of the Association of Professional Engineers of Trinidad and Tobago Vol.38, No.1, October 2009, pp.6-15 Advancing the Use of Earth Observation Systems for the Assessment of Sustainable Development Raid Al-Tahira , Terri Richardsonb and Ron Mahabirc Department of Surveying and Land Information, The University of the West Indies St Augustine Campus, Trinidad and Tobago, West Indies a E-mail: [email protected] b E-mail: [email protected] c E-mail: [email protected] Corresponding Author (Received 15 May 2009; Revised 17 July 2009; Accepted 1 October 2009) Abstract: Decisions made on the use of land in Trinidad and Tobago, with little considerations to environmental impact or physical constraints, have resulted in physical, socio-economic, and environmental problems. As a result of the country’s economic progress, urbanisation and development are fragmenting natural areas and reducing the viability of the environment to support the population. Spatial information is a crucial component in the characterisation and examination of the spatio-temporal dynamics and the consequences of the interaction between human and the environment. This information is of critical importance in the development of models to predict future trends in land cover change and therein, best land use practices to be implemented. However, the lack of data at appropriate scales has made it difficult to accurately examine the land use/cover patterns in the country. This paper argues that the gap in data and information can be managed through the adoption of earth observation technology. Moreover, it reports on the developed methodology, and highlights key results of examining the use of geo-spatial images in addressing sustainability issues associated with development. The developed methodology involves several critical steps in using multi-spectral imagery including cloud and cloud shadow removal, image classification and image fusion. Additionally, a method for improving classification performance using high resolution imagery is discussed. The results demonstrated the accuracy, flexibility and cost-effectiveness of these technologies for mapping the land cover and producing other environmental measures and indicators. Further, these results confirmed the effectiveness of this technology in establishing the necessary baseline and support information for sustainable development in the Caribbean region. Keywords: Earth Observation Systems, Spectral Image Analysis, Image Segmentation, Sustainable Development 1. Introduction The Earth’’s surface has been under constant change throughout the years. These changes have been mainly the result of anthropogenic forces in the environment. Compared to natural factors, humans pose a greater threat due to their inability to sustainably use and manage land. This has transcended into rapidly changing ecosystems largely to meet human’’s growing demands for food, freshwater, and timber (WRI, 2005). Over the years, these demands have since increased due to a variety of pressing factors facing nations worldwide, including accelerated population growth, urbanisation, migration and economic growth. These pressures placed on ecosystems are further exacerbated by issues of climate change, loss of biodiversity, growing water scarcity, and inappropriate technology applications (FAO, 2008). Specific to Trinidad and Tobago, the country has witnessed remarkable expansion, growth and developmental activities, such as building, road construction, deforestation and many other anthropogenic activities since the country’’s first oil boom in the 1940s. This has resulted in increased land utilisation, modification and alterations to the land use/cover over the years.
- 2. R. Al-Tahir etal.: Advancing the Use of Earth Observation Systems 7 Thus, a matter of grave concern is the unsustainable patterns of consumption and production that are considered the major causes for the deterioration of the environment. Development cannot survive upon a deteriorating environmental resource base and the environment cannot be protected when growth leaves out the costs of environmental destruction. Consequently, the approach of ““sustainable development”” has evolved to meet the major needs of the present without endangering subsequent needs and aspirations of future generations allowing for the conservation of nature (Gotlieb, 1996). To promote sustainability, it has become increasingly important to be able to measure how significantly vulnerable each human, environmental, and economic aspect is to damage and to identify ways of building resilience. As such, there is a need to pinpoint and implement indicators that collectively measure the capacity to meet present and future needs. The purpose of the sustainability indicators is to provide information on the state of human, environmental and economic conditions, the trend of changes in these conditions, and to identify issues that need to be addressed within each of these three pillars of sustainability (Bell and Morse, 2003). The success of any sustainability indicator depends largely on how accurately it measures reality. This depends on the use of current and accurate spatial land information, chiefly, land use and cover. In contrast, there is a severe shortage of reliable and compatible data sets in the Caribbean region. In the case of Trinidad and Tobago, this has resulted in some of the most critical datasets on the island, including land cover, being over 30 years old (Baban et al., 2004). During all these years, the land use/cover in the country would have undergone extensive change, after that map was produced. Besides being late in its delivery to represent current land, this dataset was also mapped at a scale of 1:150,000, offering a much generalised view of the land cover at the time. The present land cover dataset is not suitable for making sound decisions concerning the present and future use of land resources in the country. It is therefore necessary to adopt more effective techniques for gathering relevant spatial information to avoid problems associated with sustainable development. This is especially needed as greater land use and land cover changes will occur with the country’’s initiative of acquiring first or developed world status by the year 2020. This paper argues that the gap in data and information can be managed through the adoption of earth observation technology. Remotely sensed geo-spatial images have great potential in overcoming the information void in the country. They are relatively inexpensive and have the ability to provide information crucial to sustainable development. In this study, earth observation images were used to fill the gap in the knowledge on the state of land use and cover in Trinidad and Tobago. The objective was to undertake a detailed, spatially explicit inventory of local trends in land use and cover changes and to build a time series of land use and cover maps in order to evaluate the changes and to determine the driving forces responsible for these changes. This data could be coupled with other socio-economic and demographic data in an interdisciplinary assortment of scientific methods to investigate the causes and consequences of land use/cover change across a range of spatial and temporal scales. This paper, additionally, highlights the needs and discusses the means for the extraction of information from high resolution imagery to both support current and ongoing land cover research, and to overcome some of the present problems encountered by traditional use of medium resolution remotely sensed imagery. 2. Land Use, Land Cover and Changes Land use refers to the human activity or economic function associated with a specific piece of land (Lillesand et al., 2004). Examples of land use include agriculture, urban development, grazing, logging, and mining. Land cover, on the other hand, refers to the observed bio-physical cover on the Earth's surface (Meyer, 1995). It includes aspects of the natural environment (such as forests, wetlands, bare soil, and inland water surfaces), as well as human- made features and physical structures (such as, roads and buildings). The land use/cover pattern of a region is an outcome of natural and socio-economic factors. However, land cover today is altered worldwide primarily by direct human use: agriculture and livestock raising, forest harvesting and management, and urban and suburban construction and development. There are also indirect impacts on land cover from human activities, such as forests and lakes damaged by acid rain from fossil fuel combustion (Meyer, 1995).
- 3. R. Al-Tahir etal.: Advancing the Use of Earth Observation Systems 8 The changes in the environment brought in by anthropogenic forces have resulted in an observable pattern in the land use/cover over time. Consequently, reliable spatial and temporal information on land use/cover is critical to sustainable development. Such information serves to monitor changes on land and to understand the dynamics of those changes, leading to better planning and implementation of land use schemes. Furthermore, time series analysis of land use/cover change and the identification of the driving forces responsible for these changes are needed for the sustainable management of natural resources and also for determining the future of land use. 3. Earth Observation Systems (EOS) Observations of the earth from space provide objective information of human utilisation of the landscape. This is advantageous for monitoring and understanding the influence of human activities on natural resource bases over time. The collection of remotely sensed data facilitates the synoptic analyses of Earth-system function. This makes the detection of change possible at local, regional and global scales over time. This information is of critical importance in the development of models to predict future trends in land cover change and therein, best land use practices to be implemented. Remote sensing of the environment involves measuring electromagnetic radiation reflected from or emitted by the Earth’’s surface and relating these measurements to land cover categories or other possible surrogate indicators for environmental health (Al-Tahir et al., 2006). A variety of sensing instruments can be used to measure and record this radiation depending on its wavelength. The most commonly-used sensors are aircraft-borne cameras and multi-spectral sensors mounted on satellites orbiting the Earth. Photogrammetry has often referred to techniques for extracting information from aerial or terrestrial images, while remote sensing deals with processing multi-spectral satellite imagery. Geo-imaging techniques offer various advantages: extensive coverage, reliable and current data, and cost efficiency. Besides, they provide a unique opportunity to study the impact of land-use changes as a dynamic process across space and time, and provide proactive solutions to environmental spatial issues. In most instances, aerial or satellite imagery provides the most up to date source of data available, hence, helping to ensure accurate and reliable decisions (Al-Tahir et al., 2006). 3.1 Digital Aerial Cameras The field of photogrammetry is rapidly changing with new technologies and protocols being developed constantly. In a relatively short period of time, the practice of aerial photography and photogrammetry has gone from the analogue to digital with the advent of computing and imaging technology. The main driving motivation in developing digital photogrammetry has been the premise that it would enhance the performance and increase automation and accuracy in extracting geo- spatial information (Al-Tahir and Singhroy, 2008). One of the most obvious requirements for digital photogrammetry is concerned with the digital images themselves. While these may be obtained by scanning aerial photographs, the emerging trend is the use of digital airborne cameras for direct capturing of digital images. The first commercial digital aerial cameras were presented in 2000; nine companies now manufacture digital aerial cameras. The basic architectures are either to place linear Charge Coupled Device (CCD) arrays in the focal plane (using single lens head) or to use several area CCD chips in several cones (up to 8). A CCD is a silicon integrated circuit that enables the transportation of analog signals (electric charges) through successive stages (capacitors). Based on camera architecture, number of lenses, and intended use and applications, these cameras produce images with a dynamic range of 12 to 16 bits and an image size from 22 to over 100 megapixels (Lemmens, 2008). The new digital cameras combine photogrammetric positional accuracy with multispectral capabilities for image analysis and interpretation. Capturing of colour or multi-spectral images is achieved through adding a beam-splitter or additional lenses depending on the camera architecture. Coupled with differential GPS and inertial navigation systems (INS), these sensors generate directly georeferenced multispectral image data of any user-defined resolution up to 0.1m ground sampling distance. 3.2 High-Resolution Satellite Remote Sensing Traditionally, the extraction of information from satellite images has depended on multispectral systems, which collect data at several discrete bandwidths within the visible and infrared regions of
- 4. R. Al-Tahir etal.: Advancing the Use of Earth Observation Systems 9 the electromagnetic spectrum. As such, remote sensing based data collection has been predominantly founded on using medium resolution satellite imagery. Three platforms are currently in orbit and obtaining data; the US Landsat, the French Spot, and the Indian IRS programs. All three systems have a swath width of 60-180 km and produce multispectral data in the visible, near infrared, and short-wave infrared (SWIR) with a ground resolution of 10 to 30 m. All of these instruments have been built and operated through government-sponsored programs. Since the late nineties, private satellite corporations started collecting high-resolution remote sensing data. The satellites from GeoEye (Ikonos, launched in 1999; and GeoEye-1, launched in 2008) and Digital Globe (QuickBird, launched in 2001) are already in orbit capturing imagery at up to 0.50m ground resolution. These systems share several common specifications with respect to the spectral (number and range of spectral bands) spatial resolutions as well as orbital details. Besides, the new satellite images are recorded with 11-bit dynamic range extending the pixel values to 2048 grey shades. These new capabilities have made the use of high resolution imagery a much needed resource in a growing number of applications worldwide (Al-Tahir and Singhroy, 2008). 4. EOS in the Assessment of Sustainable Development Conventional methods of land use/cover mapping are labour intensive, time consuming and are done relatively infrequently. These maps quickly become outdated, particularly in rapidly changing environments, making monitoring and analysing change quite difficult. On the other hand, Earth observations from satellite sensors provide repetitive and spatially explicit measurements of biophysical surface attributes. As such, remote sensing has become an important source for land use/cover change assessment. Recent advances in this technology also suggest that these systems have even greater potential for providing and updating spatial information in a timely and cost-effective manner (Al-Tahir et al., 2006). 4.1 Classification of Multispectral Satellite Images Procedures for mapping land use and cover from satellite images rely heavily on the differences in spectral characteristics of the landscape for separation into land use and cover classes (Lillesand et al., 2004). Many land cover classification schemes have been developed using moderate resolution images (i.e., 20 to 250 meters ground sampling distance) in the optical and thermal wavelengths. Within this resolution range, imaging sensors smooth out variations across the individual pixels making this approach effective for use in the creation of land cover maps. After image acquisition, the process for extracting land cover information from multi- spectral images goes through the stages of pre- processing, classification and accuracy assessment before generating the final map. The pre-processing stage is necessary to restore the imagery and rectify errors and discrepancies caused by problems associated with the sensors and the platforms. This task comprises several procedures and algorithms that are often grouped into radiometric and geometric corrections (Lillesand et al., 2004). The actual extraction of distinct land use and cover categories or classes from satellite imagery is achieved at the stage of image classification. The intent of the classification process is to identify spectral signatures for the various objects on the earth’’s surface and to associate each signature with a unique land cover class. Various automated and semi-automated methods of classification do exist, the most common of which classify imagery using per pixel classifiers. Two main classification schemes exist: supervised and unsupervised classification. The essential difference between both methods lies in whether or not intervention is needed by the image analyst. Supervised classification requires such intervention, and is usually the more accurate method. The image analyst defines on the image training sites that are representative of each desired land cover category. Based on statistical analysis of the training sites, spectral signatures for each land cover category will be defined by the software and used to classify the remaining pixels. Unsupervised classification on the other hand is a fully automated process, by which the image pixels are classified by aggregating them into natural spectral grouping, or clusters (Lillesand et al., 2004). The final stage of accuracy assessment is to compare the classified imagery against ground truth (field samples). This is an important stage as the success of extracting information from remotely sensed imagery is affected by the complexity of the landscape being observed, selected remotely sensed
- 5. R. Al-Tahir etal.: Advancing the Use of Earth Observation Systems 10 data, and image-processing and classification approaches used (Lu and Weng, 2007). 4.2 Extraction of Information from High Resolution Images Within recent years, there has been increased availability and wide use of high resolution imagery in land applications. High resolution imagery shows object information such as structure, texture and detail clearly, making it ideal for observing object detail changes on the earth surface and for monitoring the extent to which humans have altered the environment (Su and Hu, 2004). It has been suggested that the application of traditional per pixel classification methods has limited applicability to high spatial resolution data because they cannot fully exploit its content (Hester et al., 2008). This is especially problematic in heterogeneous environments where pixel values are near similar and is in part due to the limited number of spectral information in such imagery (Cots-Folch et al., 2007). Additionally, high resolution imagery can be affected to a great deal by artefacts, such as shadows, making previous approaches to image classification unsuitable for this type of imagery. Image complexity and large data volumes are other general issues associated with the use of high resolution imagery that have been reported (Hester et al., 2008). One way of increasing classification accuracy with high resolution imagery is by using approaches that utilise the texture in the image. Texture is a repeated variation of intensity and colour that is directly portraying object structure and space arrangement in the image (Su and Hu, 2004). Research incorporating the use of texture measures to improve spectral classification accuracy of land cover has already met positive results (Palubinskas et al., 1995; Franklin et al., 2000; Puissant et al., 2005). A review of some of these approaches has been highlighted in (Cots-Folch et al., 2007). Those strategies targeted towards high resolution imagery include examples presented in De Martino et al. (2004) using a partial classification method in the detection of objects in an urban part of Brazil using 4m Ikonos, and Ettarid et al. (2008) in which an automated method for extracting building from 2.5m Spot and Quickbird imagery for the cities Benir Amir and Rabat in Morocco was used. Several methods use texture for image segmentation and classification; they differ mainly by the degree of prior information they require and the way texture measures are applied. Commonly used methods use statistical approaches. These are based on the measurement of the occurrences of each grey level value in a particular neighbourhood (Grey Level Co-occurrence Matrix) (Cots-Folch et al., 2007). Haralick et al. (1973) suggested a set of several features, which can be used to classify texture images; angular second moment, contrast, correlation, sum of squares, inverse difference moment, sum average, sum variance, sum entropy, entropy, difference of variance, difference of entropy, information measure of correlation 1 and information measure of correlation 2. Subgroups of these features have been widely used in research for a wide array of image classification studies. Other texture-based methods are embedded in other schema such as artificial neural networks (ANNs) and fuzzy classifiers (Shah and Gandhi, 2004; Cots-Folch et al., 2007). The ANN structure is based on the human brain’’s biological neural processes. Interrelationships of variables that are correlated in the image symbolically represent the interconnected processing of neurons of the human brain used to develop models. With fuzzy classification, there are no hard boundaries dividing geographic objects. Fuzzy classification methods assign gradual membership of pixels to classes as degrees in [0, 1], giving the flexibility to represent pixels that belong to more than one class. A review of some these applications can be found in Smits and Annoni (1999). 5. Developing the Methodology In its approach to assess the sustainability of the development in Trinidad and Tobago, this study has developed a methodology to quantify and analyse the interaction between natural and urban development in Trinidad and Tobago. As outlined in Figure 1, the methodology’’s main thrust is the use of a series of satellite images covering the period from the 1970’’s to present. These images were acquired and analysed to depict the nature of land use/cover during different times. Land cover information would be extracted from these images and combined with other available environmental, demographic, and economical data in order to define a set of mainly spatially-based sustainability indicators. A specific set of indicators; namely land use and settlement patterns, vegetation cover, loss cover, and fragmentation of land and habitat are purposely chosen because they can be extracted and updated, directly or indirectly, using
- 6. R. Al-Tahir etal.: Advancing the Use of Earth Observation Systems 11 5.1 Multi-spectral Classificationgeo-imaging techniques (Richardson and Al Tahir, 2008). The island of Tobago was chosen as a pilot study for the implementation of this methodology. The findings of this pilot are hoped to identify and address hurdles and pitfalls in the methodology. The multi-date data used to extract land use/cover information for the island of Tobago consist of four archive remote sensing images; Landsat 5 Thematic Mapper (TM) for the 1991, and Landsat 7 Enhanced Thematic Mapper Plus (ETM+) for the years 2000, 2001 and, 2002. All these images share the same 30m spatial resolution and other spectral and radiometric characteristics. The choice of these images was done on the bases of availability and the low percentage of cloud coverage. Figure 1. Methodology for Assessing the Sustainability Prior to image processing and classification of the imagery, extensive field survey was carried out within the study area to identify ground truth data for each land use/cover class sought in the classification. Some of these ground data will be used to create training sites for use in signature generation. This task is then followed by several other tasks. Figure 2 shows the workflow for the development of the land cover map based on multi-spectral data. Some details are also presented in the following sections. of Development The chosen indicators best represent the magnitude of land use/cover changes and the threats on the stability and resilience of the ecosystem. It is expected that temporal and spatial analysis of changes in these indicators against land management and physical development policies and practices would provide recommendations into the most appropriate scenario for sustainable development. Figure 2. Methodology for the Development of the Land Cover Map Using Landsat Imagery Using the image processing software Idrisi Andes (Clark Labs, Worchester, USA), the Landsat images were first geo-referenced using a total of ten ground control points extracted from the 1:25000 topographic maps of Tobago. Images were then atmospherically corrected, using the dark object subtraction model (Lillesand et al., 2004). One disadvantage of optical imagery in tropical environments, more specific to the Caribbean and Trinidad and Tobago, is the persistent cloud cover that complicates the processing of satellite imagery. Pixels represent clouds and their shadows in the scene must first be pinpointed and masked using one approach or another. The method adopted for this research was a semi-automated approach that was put forward by Martinuzzi et al. (2003) and modified by the authors to produce a new cloud and cloud- shadow masking technique. The method involves identifying contaminated pixels and developing a mask using Landsat image values in the blue wavelength (band 1) and in thermal range (band 6.1).
- 7. R. Al-Tahir etal.: Advancing the Use of Earth Observation Systems 12 Initially, an unsupervised classification was performed to identify patterns of general spectral categories relating to the land cover. Subsequently, a supervised classification was performed on the Landsat TM and ETM+ images for 1991 and 2002. Training sites selection was guided by the cover types identified during the unsupervised classification and a priori knowledge of the study site. A signature file representing each individual land cover class was created by the software and used to classify the remaining image using the Maximum Likelihood Classification method (Lillesand et al., 2004). These classes included, forest, savannah and agriculture, urban and water (sea). The final classified image contained large data gaps as a result of the removal of cloud and cloud shadow pixels. These gaps were filled using photo interpretation techniques and knowledge of the study site. This was done using other available higher resolution imagery for Tobago, including 2003 Ikonos imagery (1m resolution colour image) and the 1994 mosaic of panchromatic aerial photographs. The land cover map is also updated for other features in the image that were not distinguished by the supervised classification process for different reasons (e.g., cloud cover, spatial resolution of images). The land water bodies of the Pigeon Point, Kilgwyn swamp, and Hillsborough dam are examples for such missing features on the final classification output. These features were identified on the high resolution imagery, digitised on-screen, and finally used to update the classified image. Figures 3 and 4 show the completed land cover maps for Tobago for 1991 and 2002, respectively. An accuracy assessment was performed for the 1991 and 2002 derived land cover maps by comparing these results with reference ground truth data (151 sample sites). The accuracy was derived by means of error matrix (confusion matrix), which calculated the overall accuracy to be 89.4% for the classification of 1991 image, and 88.7% for the classified image of 2002. While they can be slightly improved, these accuracy values were considered appropriate at this stage in the research. 5.2 Non-spectral Image Segmentation In the Caribbean, there are large archives of panchromatic aerial photographs dating back to the middle of the last century. The use of this data will permit access to historical information, critical for any temporal analysis of land cover. Additionally, aerial photographs do not suffer from cloud cover effects. However, multispectral classification techniques cannot be applied to panchromatic aerial photographs. Based on available literature, there has not been any attempt for an automated approach to mapping and monitoring land cover information in the country from aerial photographs. Figure 3. The Land Cover Map of Tobago for 1991 Figure 4. The Land Cover Map of Tobago for 2002 At present, the authors have embarked upon a research effort to utilise texture measures for improving the current performance of presently used classification methods in the region. This research is still in its preliminary stages but from an extensive survey of the literature to date a proposed methodology has already been purported based on textural measures from aerial photographs (Figure 5). This approach utilises texture measures derived from the Grey Level Co-occurrence Matrix (GLCM) for various size windows along with varying co- occurrence pixel angles in the image. A supervised classification technique is then used to extract object detail to a resulting land cover map. 6. Conclusion Trinidad and Tobago can be characterised as small islands with fast rates of development.
- 8. R. Al-Tahir etal.: Advancing the Use of Earth Observation Systems 13 Figure 5. Methodology for Land Cover Mapping Using Image Texture Many of the farmlands, forests, and wetlands have been transformed at unprecedented rates into human settlements. Thus, there is a growing concern about the untenable patterns of urban sprawl, loss of natural vegetation and open space, and a general decline in the health of the environment. To achieve sustainability, there is a need to measure the changes in land use/cover that have occurred and to predict the impact of future changes in order to identify the factors that cause deterioration of the environment. One significant source of current and reliable geographic information on land use and cover is air and space borne imaging sensors. Images from earth observing systems have an important role to play in maintaining the equilibrium between the sustainable management of natural resources, environmental protection and rapidly increasing population. It is the view of the authors that the effective use of remote sensing data and a suitable blend with environmental and socio-economic data would help in achieving a local specific prescription to realise sustainable development in the Caribbean region. The pilot study in Tobago has produced rewarding results and demonstrated the flexibility and cost-effectiveness of these technologies for mapping the land cover. Other environmental measures and indicators can also be derived from this data and augment the analysis. There are limitations in using satellite images for monitoring land use/cover changes. Firstly, the medium resolution of the imagery impacts on the size of features that can be distinguished in the image. This may affect the accuracy of image classification and assessment of changes over time. The second limitation is the unavailability of multi- date images of satisfactory quality, especially with respect to a low percentage of cloud cover of the image. Aerial photographs, as well as high resolution satellite images, most definitely provide a valid alternative. This is especially beneficial since there exist an available achieve of aerial photographs spanning back to the middle of the last century. With this information a more accurate depiction and a larger time span of land cover changes can be studied. However, a robust approach for extracting information in format and scale compatible to those of the satellite images has not yet been developed. Consequently, the study has embarked on developing a methodology for extracting land use/cover information based on texture and tone in the image rather than the spectral components. The developed methodology is expected to provide a faster approach for updating current and future land cover maps of the country. Other Caribbean islands or other countries with similar settings can also adopt this methodology and expect similar or greater benefits. References: Al-Tahir, R. and Singhroy, V. (2008), ““Mapping landslides in tropical environment using contemporary geo-imaging technologies”” in Baban, S. (ed.). Enduring Geohazards in The Caribbean: Moving from the
- 9. R. Al-Tahir etal.: Advancing the Use of Earth Observation Systems 14 Reactive to the Proactive, The University of West Indies Press, Jamaica, Chapter 5, p.81-103 Al-Tahir, R., Baban, S., and Ramlal, B. (2006), ““Utilising emerging geo-imaging technologies for the management of tropical coastal environments””, The West Indian Journal of Engineering, Vol.29, No.1, pp.11-21 Baban, S., Ramlal, B., and Al-Tahir, R. (2004), ““Issues in information poverty and decision making in the Caribbean region: A way forward””, The West Indian Journal of Engineering, Vol.27, No.1, pp.28-37. Bell, S. and Morse, S. (2003), Measuring Sustainability; Learning by Doing, London: Earthscan Publications. Cots-Folch, R., Aitkenhead, M.J. and Martinez- Casasnovas, J.A. (2007), ““Mapping land cover from detailed aerial photography data using textural and neural network analysis””, International Journal of Remote Sensing, Vol.28, No.7-8, pp.1625-1642. De Martino, M., Macchiavello, G. and Serpico, S.B. (2004), ““Partially supervised classification of optical high spatial resolution images in urban environment using spectral and textural information””, Proceedings of IEEE International Geoscience and Remote Sensing Symposium, Alaska, Vol.1, pp 80-91 Ettarid, M. Rouchdi, M. and Labouab (2008), ““Automatic extraction of buildings from high resolution satellite images””, The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol. XXXVII, Part B8. Beijing, pp. 61-66 FAO (2008), Feeding the World Sustainable Management of Natural Resources, Food and Agriculture Organisation, United Nations, Rome; www.fao.org/docrep/010/ai549e/ai549e00.htm Franklin, S.E, Hall, R.J., Moskal, L.M., Maudie, A.J. and Lavigne, M.B. (1990), ““Incorporating texture into classification of forest species composition from airborne multispectral images””, International Journal of Remote Sensing, Vol.21, No.1, pp.61––79. Gotlieb, Y. (1996), Development, Environment and Global Dysfunction, Towards Sustainable Recovery, Florida: St. Lucie Press. Haralick, R.M., Shanmugan, K. and Dinstein, I. (1973), ““Texture features for image classification””, IEEE Transactions on System, Man and Cybernetics, Vol.3, No.6, pp.610-622. Hester, D.B., Cakir, H.I., Nelson, S.A.C. and Khorram, S. (2008), ““Per-pixel classification of high spatial resolution satellite imagery for urban land-cover mapping””, Photogrammetric Engineering and Remote Sensing, Vol.74, No.4, pp.463-471. Lemmens, M. (2008), ““Digital Aerial Cameras - Product survey””, GIM International, Vol.22, No.4, pp.22-25. Lillesand, T., Kiefer, R., and Chipman, J. (2004), Remote Sensing and Image Interpretation, 5th edition, New York: John Wiley & Sons. Lu, D. and Weng, Q. (2007), ““A survey of image classification methods and techniques for improving classification performance””, International Journal of Remote Sensing, Vol.28, No.5, pp.823-870. Martinuzzi, S., Gould, W. and Ramos, O. (2003), ““Cloud and cloud-shadow removal in the creation of a cloud free composite Landsat ETM+ scene in tropical landscapes””, Presented at the National GAP Annual Meeting, Fort Collins, Colorado, USA. Meyer, W.B. (1995), ““Past and present land-use and land- cover in the USA””, Consequences, Vol.1, No.1, pp.24- 33. Palubinskas, G., Lucas, R.M., Foody, G.M. and Curran, P.J. (1995), ““An evaluation of fuzzy and texture-based classification approaches for mapping regenerating tropical forest classes from Landsat-TM data””, International Journal of Remote Sensing, Vol.16, No.4, pp.747––759. Puissant, A., Hirsch, J. and Weber, C. (2005), ““The utility of texture to analysis to improve per-pixel classification for high to very high spatial resolution imagery””, International Journal of Remote Sensing, Vol. 26, No.4, pp.733-745. Richardson T. and Al-Tahir, R. (2008), ““Modelling land use and land cover dynamics to assess sustainability in Trinidad and Tobago””, Proceedings of the 10th International Conference for Spatial Data Infrastructure, GSDI Association.Trinidad and Tobago. 15 pages. Shah, S.K. and Gandhi, V. (2004), ““Image classification based on textural features using artificial neural network (ANN)””, Electronics and Telecom Engineering, Vol. 87, pp.72-77. Smits, C.P. and Annoni, A. (1999), ““Updating land-cover maps by using texture information from very high- resolution space-borne imagery””, IEEE Transactions on Geoscience and Remote Sensing, Vol.37, No.3, pp. 1244-1254. Su, Junying and Hu, Qingwu (2004), ““Fast residential area extraction algorithm in high resolution remote sensing image based on texture analysis””, Istanbul, ISPRS. Available from Internet: www.isprs.org/istanbul2004/comm7/papers/214.pdf (Last accessed on July 25, 2008) WRI (2005), World Resources 2005: The Wealth of the Poor-Managing Ecosystems to Fight Poverty, In collaboration with United Nations Development Programme, United Nations Environment Programme, and World Bank. Washington, DC: World Resources Institute Biographical Notes: Raid Al-Tahir is the coordinator for the Centre for Caribbean Land and Environmental Appraisal Research (CLEAR), and a Senior Lecturer in Photogrammetry and Remote Sensing in the Department of Surveying and Land Information, at the University of the West Indies, St Augustine, Trinidad and Tobago. He received a BSc in
- 10. R. Al-Tahir etal.: Advancing the Use of Earth Observation Systems 15 Surveying Engineering from the University of Baghdad (Iraq) in 1980, and MSc and PhD from The Ohio State University (USA) in 1989 and 1995, respectively. His research interests are in the areas of environmental geoinformatics and algorithmic aspects of processing geo-spatial images. Terri Richardson received a BSc in Surveying and Land Information from the University of the West Indies. She is currently a Research Student and a Graduate Research Assistant in the Department of Surveying and Land Information. Her research interests include the use of remote sensing for mapping and assessing the changes in land use and land cover and their relations with sustainable development. She has received in 2008 The Commonwealth Association of Surveying and Land Economy (CASLE) Award for Young Authors. She is a member of the Centre for Caribbean Land and Environmental Appraisal Research (CLEAR), UWI. Ron Mahabir received his BSc in Computing and Information Systems from the University of London and his MSc Geoinformatics from The University of the West Indies in 2004 and 2008 respectively. He is currently an Assistant Lecturer in the Department of Surveying and Land Information, University of the West Indies, and working towards a PhD degree in Geoinformatics in the same university. He is a member of the Centre for Caribbean Land and Environmental Appraisal Research (CLEAR), University of the West Indies and a member of the International Society for Photogrammetry and Remote Sensing Student Consortium. His current research interests include feature extraction from high resolution imagery, pattern recognition, computer vision, and image analysis.