REMOTE SENSING AND IMAGE INTERPRETATION CHANGE ...

UNIVERSITA’ DEGLI STUDI DI FIRENZE

REMOTE SENSING AND IMAGE INTERPRETATION CHANGE DETECTION ANALYSIS: CASE STUDY OF BORGO PANIGALE AND RENO DISTRICTS

Master II Livello in Sistemi Informativi Geografici per il monitoraggio e la gestione del territorio

Candidato: Antonio Tabarroni AA 2009/10

CONTENTS

1. INTRODUCTION ........................................................................................................................................................ 1 2. METHODOLOGY ....................................................................................................................................................... 4 3. STUDY AREA ............................................................................................................................................................... 5 4. IMAGE ACQUISITION AND SPECTRAL PROPERTIES ..................................................................................... 7 4.1 IMAGE ACQUISITION AND BANDS COMBINATION ........................................................................................................ 7 4.2 SPECTRAL RESPONSE OF VEGETATION ....................................................................................................................... 9 4.3 AGEA ORTOPHOTOS (2008) AND QUICKBIRD SATELLITE IMAGES(2003) ................................................................ 10 4.4 ENVI PROCEDURES: BASIC TOOLS ......................................................................................................................... 11 5. CLASSIFICATION METHODS: SUPERVISED CLASSIFICACION AND TRAINING DATA ...................... 14 5.1 IMAGE INTERPRETATION: SUPERVISED CLASSIFICATION .......................................................................................... 14 5.2 REGION OF INTEREST (ROI) AND TRAINING DATA ................................................................................................. 17 5.2.1 ENVI Procedure – Region of Interests............................................................................................................ 19 5.3 SAM(SPECTRAL ANGLE MAPPER ) ......................................................................................................................... 20 5.3.1 ENVI Procedure – SAM.................................................................................................................................. 21 6. POST CLASSIFICATION ......................................................................................................................................... 22 6.1 CONFUSION MATRIX ................................................................................................................................................ 22 6.1.1 ENVI Procedures: how to calculate Confusion Matrix .................................................................................. 23 6.2 DATA MANAGEMENT: LAND USE OVERLAP ............................................................................................................. 24 6.2.1 ArcGis Procedures.......................................................................................................................................... 25 6.3 DECISION TREE ....................................................................................................................................................... 25 6.3.1 Decision Tree procedures ............................................................................................................................... 26 7. A CHANGE DETECTION CASE STUDY: RENO AND BORGO PANIGALE DISTRICTS (2003-2008) ....... 27 7.1 STUDY AREA ........................................................................................................................................................... 28 7.2 ENVI PROCEDURES: CHANGE DETECTION .............................................................................................................. 29 8. CONCLUSION ........................................................................................................................................................... 30 9. FIGURES .................................................................................................................................................................... 32 10. REFERENCES ......................................................................................................................................................... 55

1. INTRODUCTION The thesis will discuss the analysis of urban green areas in the Municipality of Bologna and more specifically change detection of land use in precise districts such as Reno and Borgo Panigale. The analysis is conducted during training and internship activities at Arpa Remote Sensing department of Emilia-Romagna. Arpa stands for “Agenzia Regionale per la Prevenzione e l´Ambiente Emilia-Romagna” (Regional Agency for Environmental Protection in Emilia-Romagna)1 established under Regional Law n.44 of 19-4-1995 and subsequent amendments.

It is an environmental control technical support body to the regional, district and local authorities and is administratively and technically independent. Arpa´s functions, activities and tasks cover all aspects concerning environmental control, including:  monitoring of the various environmental components  management and surveillance of human activities and their territorial impacts  activities in support of the environmental impact assessment of plans and projects  creation and management of a regional environmental information system.

The present work shows how to use multi spectral images for monitoring vegetation extent in the urban environment of the metropolitan area of Bologna. This study aims to map urban green areas (quantitative method) and main vegetation types as Needle-Leaves, Broadleaves, different kind of grass and meadows (qualitative method) using remote sensing analysis.

Cities are places where it accentuates the maximum human pressure and where the therefore focus on more than environmental imbalances that cause the greatest damage to physical and mental health of citizens. For this reason the study of the environmental quality into metropolitan areas, aimed at identifying effective measures for its improvement is extremely important. A natural component that can influence the quality of the environment and city life consists of its urban vegetation, which, in addition to the aesthetic and recreational functions, helps to mitigate the

1

http://www.arpa.emr.it/cms3/documenti/_cerca_doc/istituzionali/arpa_brochure_eng.pdf

1

pollution of the environmental urban and increase biodiversity.

However, the lack of a uniform system of green's detection, mainly because of the absence monitoring of a specific part of many municipalities and the absence of an exact definition of indicators for the analysis of urban green spaces, making it difficult to compare the state of the city in a consistent analysis. For this reason it is necessary to bridge the information gap and consequently is essential to establish instruments for detecting the green space on which to base a scientific approach to the study of its effects.

The term 'urban green' are associated with different definitions, especially depending on the context in which is treated. According to ISTAT sources, urban green spaces term is defined as “the wealth of green spaces which insists maintained on the municipalities, directly or indirectly, by public bodies such as municipalities, provinces, regions and the state. In this context include different types of parks: green spaces, urban parks, historical gardens, urban areas and special areas, including gardens school, botanical gardens, nurseries, zoos and other residual categories” (ISTAT, 2001)2.

The inventory of urban green spaces is therefore an important tool to use planning of cities, both for environmental and for the design of new areas, and also has a qualitative value regarding the presence of constraints or the need for maintenance.

Finally, the management of urban green areas is one of the basic elements to ensure efficient policy sector and especially a more rational use of soil resources. In this case I shall consider a green urban areas as open vegetation (Needle-Leaves or Broadleaves, sparse grass, meadows) and closed vegetation (trees, foliage, shrubs, forest areas) plus spectral response by asphalt, hydrology, and urbanized industrial areas. The transformation and development of green spaces in the town of Bologna were addressed by comparing images AGEA (ortophotos) for the year 2008 and QuickBird satellites images for the year 2003 and specific land use classification.

The preparation of the map data requires use of a variety of types of image sources. For this work I have analysed AGEA year 2008 orthophotographs with RGB and CIR band combination and similar classification obtained by land use analysis thanks to Arpa colleagues

2

http://www.istat.it/salastampa/comunicati/non_calendario/20080828_00/testointegrale20080828.pdf

2

during 2003. The opportunity to record the areas of urban green spaces and transformations of change detection (especially from vegetation use to urbanize) is possible by using specifics remote sensing software: it was decided to exploit the potential of Envi 4.7. and ITT suites.

3

2. METHODOLOGY The methodology of the study involved: 

Creation of base layers as boundary of Bologna's Municipality



Combination of RGB and CIR AGEA Orthopohotos



Clipping images and features.



Mosaicking of data corresponding to the study area



Computation and analysis of Region of Interests (ROI)



Histogram generation, Bi-spectral plots, Regression analysis



Classification Supervised



Statistical analysis and report generation after Spectral Angle Mapper procedure (SAM)



Creation of QuickBird 2003 image layer of Bologna's Municipality



Decision Tree to compare classification



Change Detection, visualisation and assessment of change analysis.



Statistical analysis and report generation about Change Detection



Reno and Borgo Panigale districts extraction mask



Practical examples and conclusion on Reno and Borgo Panigale Change Detection

4

3. STUDY AREA Bologna is the capital city of Emilia Romagna, in the Po Valley of Northern Italy. The city lies between the Reno river, the Savena River and the Appennine Mountains. Its municipality (extended for 140.85 Kmq) include nine districts such as Navile, Porto, San Donato, Santo Stefano, San Vitale, Saragozza, Savena, Borgo Panigale and Reno (these two districts will be review during land use change detection's case study in the last chapter).

To compare statistics about population and surface it is better analyse tables on total resident and density population3. (Figure 1) We note low population density in the district of Borgo Panigale in which, since 2003, designers are focusing to build new residential buildings, especially popular buildings. Where population density is higher (as Barca districts) we note increasing residential buildings' areas such as villas that produce higher aesthetic coherence and accuracy with territory around. In Bologna’s varied and complex landscape several factors that mark the city’s identity more profoundly are immediately evident: the old town centre, the range of hills (accounting for 28% of the municipal area), the Reno and Savena rivers and the open countryside on the plain. This considerable diversity in the territory, whose overall surface is 14,087 hectares, is matched by a notable wealth of natural, semi-natural and man-made environments, equally worthy of interest in terms of territorial management and planning policies directed at guaranteeing, as required, protection, improvement and restoration.

Public green spaces in Bologna now include more than 750 areas, covering a total of over 1,000 hectares. Actual parks and public gardens total around 250 (600 hectares) added to which are the green areas created alongside roads (160 hectares), sports centres (110 hectares), green areas linked to schools, green areas around public buildings and many smaller areas as well (totalling 180 hectares)4. It is an asset of considerable size if placed in comparison with many other Italian cities. Nonetheless, it provides too few top quality areas and responds only partially to the many requirements of the inhabitants in terms of the quality of green spaces and how they are laid out.

3 4

Comune di Bologna - Dipartimento Qualità della Città - Settore Ambiente http://www.laboratoriorapu.it/Plans_Project

5

Plans to consolidate and increase Bologna’s green areas must necessarily take into account the wider territorial context. This means considering the ten municipalities bordering on the city and, to a lesser extent, other municipalities that border on them, with the aim of integrating the layout of Bologna’s green spaces with those that exist, or are under development, in the neighbouring territories. Among the various themes that have emerged, the contribution that Bologna can make to a wider project of protecting, enhancing and using the Reno river stands out. This initially involves the municipalities of Calderara di Reno, Casalecchio di Reno and Sasso Marconi, but could also extend to the municipalities located on the banks of the river from the mountain to the plain, at a later date.

Also the area around the banks of the Savena could provide a good opportunity for a project at superior municipal level, as the footpath along the banks in Bologna could be continued through the territories of San Lazzaro and Castenaso. As for the hills above Bologna, a good policy appears to be to increase links with the adjacent river environments of the Reno and the Savena, the areas protected by the Gessi Bolognesi and Calanchi dell’Abbadessa Regional Park and those belonging to the planned Nature Reserve at Contrafforte Pliocenico. The theme of safeguarding and enhancing the countryside around the urban area promotes the opportunity to create new and effective connections between the areas of greatest natural beauty in the Bologna area and the surrounding open spaces.

A further idea being developed, which could be an excellent opportunity for co-operation between Bologna and the neighbouring municipalities, is to promote the wonderful system of old canals that flow through Bologna, in particular the two most important artificial waterways: the Reno canal and the Navile canal.

6

4. IMAGE ACQUISITION AND SPECTRAL PROPERTIES The object of computer processing of image data is the extraction of useful information about the scene. Whatever its source, the image must be digitized before it can be processed by a digital computer.

For most remote sensing systems, digitization is a part of the imaging although,

occasionally, digitization is a separate step. In either case, this step is an important one since the decisions made at this level will significantly affect the information content of the digitized image. Different kinds of images such as the satellites ones and processed orthophotos have been used during this project to obtain a better result.

4.1 Image acquisition and bands combination The possibility to obtain the colours by combination of basic colours is used in Remote Sensing to represent in the screen multi spectral images. Representation pattern used as reference is RGB model (red, green, blue).

It s possible to obtain three kinds of representation in the screen:

-Black and White Image: a single special band is visualized matching the pixel values with all three basic colours, with same strength, obtaining in the screen several shades of grey.

-True Colours Image: blue colour is joined to blue spectral band, green to green spectral band, red to red spectral band (RGB,TM3,TM2,TM1) obtaining a colours combination like the one perceived by human eye for the same scene.(Figure 2)

-False Colours Image : in this scheme there is no spectral bands relation in which image has been acquired and the colour of representation. False Colour Infra-Red image (IRFC) when near IR band is represented in red, red band in green and green band in blue (like a vegetation analysis as RGB:4,3,2).This combination is frequently used: in the aerial photographs IR false colour this kind of code is used so that the vegetation owing to comparatively high reflection value in IR looks prevalently red. (Figure 3) 7

Although it's possible, having at disposal in a multi-spectral image several spectral bands any arbitrary association of band and colour of visualization, the purpose of any work is to guide such choice, the best representation shows the data of interest. The high frequency of spatial change in urban area, along with the great mixture of surface materials, produces distinctive urban spectral responses, making most of the urban categories separable in imagery. Yet, urban spectral responses can also be said to resemble familiar curves from vegetation, soil and rock or water.

Actively growing plants therefore show a strong contrast between strong absorption in the red and high reflectance in the Near Infra-Red (NIR) regions of the spectrum. The amount of absorption in the red and reflectance in the near Infra-Red varies with both the type of vegetation and the health of the plants. Normally the healthy green leaf has very low reflectance values in the red (600-700 nm) due to chlorophyll absorption and very high reflectance values in the Near Infra-Red (700-1000 nm). However, as the plant begins to turn yellow, reflectance begins to decrease in the Near InfraRed and increase in the red. The decline in reflectance in the Near Infra-Red is due to the spongy mesophyll layer collapsing as the leaf comes under stress and the increase in reflectance in the red is caused by the die-off of chlorophyll and therefore a decrease in absorption.

Relatively pure vegetation pixel occurs in few places in cities, for example parks or cemetery. “Vegetation plays a decreasing role in the other urban land use categories from low-density residential areas with much vegetated surface to other such as high-density commercial or industrial areas with almost no vegetation. The influence of vegetation on the spectral response curve is therefore similar to those obtained in natural areas.”5 The spectral responses of many urban construction materials are similar to soil responses with the absence of water absorption band. Concrete and asphalt, for example, may occur on opposite end of a soil line in a plot Infra-Red data against red data, with asphalt appearing dark soil pixels in the diagram of spectral response.

5

M. Lillesand, Remote Sensing and image interpretation, Wiley, New York, 1979 page 245

8

4.2 Spectral response of vegetation Field personnel should have a good understanding of the spectral response of vegetation, including variations in response through the growing season, and with changes in moisture content. Knowledge of these variations aids in selection of the optimal time for field work and helps in selecting biophysical features for measurement. Interaction of electromagnetic energy with leaf pigments is restricted to the visible wavelengths (400-700nm).When chlorophyll is abundant in the leaf; it dominates both reflection (in the green band) and absorption (in blue and red bands). This accounts for the two absorption bands on either side of the green reflectance band and explains why a leaf appears green. (Figure 4)

When the leaf begins to change, due to dehydratation, chlorophyll production slows or stops, and reflection in the blue and red bands is increased while reflection in the green band is reduces. If other pigments remain in the leaf, they will dominate the reflection and the leaf will change colour. Pigments such as carotene and Xanthophyll reflect primarily as yellow and browns, and Anthocyanin produces a strong red. In the visible wavelengths absorption and reflection at the leaf surface is dominant, and relatively little energy is transmitted through the leaves of most species.

The loss of chlorophyll removes the absorption bands in the blue and red, creating a smoother curve in the visible. Loss of all pigments during senescence produces a high reflectance across the visible bands. Information on the interaction of energy and plant pigments can help in planning field measurements.

A field worker familiar with spectral curves of plant species would know that some species are almost indistinguishable by spectral analysis in the visible wavelengths. Another species may have less blue absorption and produce a slightly blue-green colour, making it easily identified. Also, the low level of energy transmission through leaves in the visible wavelengths means that those wavelengths are not useful for making estimates involving plant volume, such as biomass or leaf area index. The visible wavelengths, however, serve well for estimates of plant coverage, such as percent cover or crown closure. Visible wavelengths will provide information related to moisture content only when a plant is under enough stress to stop producing chlorophyll and change colour.

9

4.3 AGEA Ortophotos (2008) and Quickbird satellite images (2003) The availability of orthophotos and aerial photography is very important for the government to update spatial data in many planning phases as environmental monitoring and controlling. Starting from the consideration that for the certification of agricultural production each year AGEA usually produces about one third of the Italian orthophotos, Emilia-Romagna has proposed a collaboration to achieve in 2008 the whole of the Emilia-Romagna boundary. Our interest is focus about stereoscopic aerial photographs and digital-pixel average of 50 cm.

The first procedure involves the preparation of 2008 AGEA orthopothos in a mosaic that includes only the municipality of Bologna as the ultimate goal plan to compare same classifications' classes about 2003 and 2008 years. AGEA orthophotos by ARPA was generated according to the following product specifications6:  Pixel size: 50 cm;  Gauss-Boaga reference system;  Radiometric depth 8 bits per band;  Spectral bands: Blue, Green, Red, Near Infra-Red;  Mapped area of each file corresponds to the ortho-CTR10 sections, or the sixteenth part of a series of paper topographic national IGM50 and characterized, similarly to the sheets of CTR10, a six-digit code in the first three are borrowed from IGM50 numerical code of the paper, the fourth and fifth progression rows / columns of its 16 breakdowns, the sixth is constantly zero and qualification in the Italian application context, the mapping at 1:10,000 scale;  Digital uncompressed TIFF + TFW;  The sections must be radiometrically homogeneous, and at the mosaic of frames for flights of different date must be made according to cutting lines below the physical limits natural land so as not to accentuate the evidence;  Clouds and mist must be less than 5% in each section is less than 2% on the EmiliaRomagna has also been specifically requested by the region including the areas with and settlements found to be substantially free of clouds. 6

Regione Emilia Romagna, Ortofoto Multifunzione ERAGEA2008, page 8

10

However, another source of data used is the QuickBird satellite images relating to year 2003. (Figure 5) QuickBird is a high-resolution commercial earth observation satellite, owned by DigitalGlobe and launched in 2001 as the first satellite in a constellation of three scheduled to be in orbit by 2008. QuickBird uses Ball Aerospace's Global Imaging System 2000 (BGIS 2000) that collects the fourth highest resolution commercial imagery of Earth after WorldView-1,WorldView-2 and GeoEye-and boasts the largest image size and the greatest on-board storage capacity of any satellite. The satellite collects panchromatic (black & white) imagery at 60-70 cm resolution and multi-spectral imagery at 2.8-meter resolutions.

At this resolution, detail such as buildings and other infrastructure are easily visible. The imagery can be imported into remote sensing image processing software, as well as into GIS packages for analysis. These images, acquired on 07.22.2003, have four bands (Blue, Green, Red and Near InfraRed) to 11 bits of resolution radiometric and geometric resolution of 2.8 meters. Each image is cut on the elements of the CTR at 1:5,000 scales. The data set was kindly made available by the Province of Bologna through a convention for the analysis of data established by the laboratory of ARPA remote sensing.

4.4 ENVI Procedures: Basic Tools The first step describe basic preparation, data subset , resizing and mosaicking operations to obtain whole Bologna's boundaries image as RGB (4,3,2) combination because external boundaries (outside municipality of Bologna) are not necessary to analyse for this case study's objectives and it surely use unnecessary storage capacity.

The descriptions are thus the phases of combination bands, clips and mosaicking. This band combination (4, 3, 2) will create a colour Infra-Red image (CIR) similar to that used in CIR photography. The combination uses the Near Infra-Red band of data in the image and vegetation will appear red. Vegetation is highly reflective in this part of the spectrum so this display is a very powerful way to examine changes in vegetation.

11

This is the combination used for urban green detection analysis during several phases that I will describe in the following paragraph:

A - From File menu choose to save file as ENVI Standard B - Select Import Builder menu clicking importing RGB from spectral subset and select only Blue band C - After that similar operations must be made as Green and Red band in this order D - Repeat the same phases about CIR band that it joins with RGB E - Set Map Info (right click on map Info-edit map information) and set the reference system. I will load the image and decide to display it in grey scale or RGB (in this case 4, 3, 2 combinations) with WGS84 Datum and UTM 32 North projection as reference system

Whole sections derived from CTR (Carta Tecnica Regionale) within the boundary of municipality of Bologna must be cut off to have spaces that flow out from the edge of town and not relevant to this case study. Many sections contain very small part of the area and therefore it is necessary to clip. (Figure 6)

Masking operations are possible before those mosaicking ones with the tools to edit images generated by using the command ENVI 4.7 as Basic Tools and applying to the mask EVF file of the town of Bologna. The EVF file extension is a file that represents the vector files in ENVI Software products. Then it is equivalent as a SHP extension but both files must be converted before use. At the moment with ENVI release there are commands able to manage ARCGIS file into ENVI.

To obtain entire mosaic clip is essential having adjacent images. A series of orthopotographs that show adjacent regions on the ground can be joined together to form a mosaic. Controlled mosaics are formed form individual images assembled in a manner that preserves correct positional relationships between the features they represents. Often the most accurate region of each photograph near the principal point is cut out and used for the mosaic. Locational control must, of course, be provided by ground survey or by information from accurate maps. Mosaicking involves combining multiple images into a single composite image.

ENVI provides interactive capabilities for placing georeferenced images within a mosaic, and automated placement of georeferenced images within a georeferenced output mosaic. ENVI also provides transparency, histogram matching, and automated colour balancing. 12

Mosaic is useful when a set of adjacent raster datasets needs to be merged into one entity and also when minimizing the abrupt changes along the boundaries of the overlapping rasters. The overlapping areas of the mosaic can be handled in several ways; for example, I can set the tool to keep only the first raster dataset, or blend the overlapping cell values. There are also several options to determine how to handle a colour map, if the raster dataset uses one, for example, keeping the colour map of the last raster dataset used in the mosaic. For mosaicking of discrete data, the First, Last, Minimum, or Maximum options will give the most meaningful results. The Blend and Mean options are best suited for continuous data.

ENVI 4.7 procedures for mosaicking operations are the following:

A - Load all the pictures from mosaic by Ctrl-Open Image File (in this case all the sectioned image) B - Use the tools Mosaic by Georeferenced MapBased window and choose Import Files and edit profiles C - Select images and click OK D - Set the values of Data Value = 0 and linear stretch = 0% for all images E - Load and accept. (Figure 7) This image represents the mosaicked mask about municipality of Bologna's boundaries. This is the base on which I have worked to get various types of classification and data analysis described in the following chapter.

13

5. CLASSIFICATION METHODS: SUPERVISED CLASSIFICACION AND TRAINING DATA Supervised classification can be used to cluster pixels in a dataset into classes corresponding to user-defined training classes. The following classification requires a precise selection of training areas for use as the basis for classification. Various comparison methods are then used to determine if a specific pixel qualifies as a class member.

5.1 Image interpretation: Supervised Classification In this plan I have followed comparatively two operative formalities in order to gain several classification patterns. The first one has followed a homogeneous classification process as carried out in 2003 to obtain a change detection analysis. The second one, instead, has followed a more precise classification to better define the vegetation condition and urban area with a wider range of items. For both I have used the same classification techniques supervised classification with spectral angle mapper method (SAM) with an essential difference during decision tree running phases. At first is better to explain the fundamental classification concepts with a short introduction in the file of image interpretation and remote sensing.

Digital image classification is the process of assign pixel to classes. Usually each pixel is treated as an individual unit composed of values in several spectral bands. By comparing pixels to one another, and to pixels of known identity, it is possible to assemble groups of similar pixels into classes that are associated with the informational categories of interest to users of remotely sensed data.

These classes form regions on a map or an image, so that after classification the digital image is presented as a mosaic of uniform parcels, each identified by a colour or symbol. These classes are, in theory, homogeneous: pixels within classes are spectrally more similar to one another than they are to pixels in other classes. In practice each class will display some diversity, as each scene will exhibit some variability within classes. “Image classification is an important part of the fields of remote sensing, image analysis, and pattern recognition. In some instances, the classification itself may be the object of the analysis”7. 7

ENVI Tutorial: Classification Methods http://www.ittvis.com

14

The simplest form of digital image classification is to consider each pixel individually, assigning it to a class base up upon its several values measured in separate spectral bands. Sometimes such classifiers are referred to as a spectral or point classifier because they consider each pixel as a 'point'.

That is why spectral classes are so important for analysis method. Spectral classes, in fact, are groups of pixels that are uniform with respect the brightness in their several spectral channels. The analyst can observe spectral classes within remotely sensed data; if it is possible to define links between the spectral classes on the image and the informational classes that are of primary interest, the images forms a valuable source of information. Thus remote sensing classification proceeds by matching spectral categories to informational categories. If the match can be made with confidence than the information is likely to be reliable. If spectral and informational categories do not correspond, then the image is unlikely to be a useful source for that particular form of information. Furthermore, other factors, such as variations in illumination and shadowing may produce additional variations even within otherwise spectrally uniform classes.

The main kind of distinction in image classification separates supervised classification from unsupervised classification. The former procedures require considerable interaction with the analyst, who must guide the classification by identifying areas on the image that are known to belong to each category. The ladder, on the other hand, proceeds with only minimal interaction with the analyst, in a search for natural groups of pixel present within the image. During planning and operative phases I decided to use supervised classification to obtain a more consistent result in line with the expectations of final goal, trying to analyse the territory and the various types of vegetation. That is why, despite many difficulties, working as supervised classification exploit many advantages that I describe below.

Supervised Classification can be defined informally as the process of using samples of known identity to classify pixel of unknown identity. Samples of known identity are those pixels located within training areas or training fields. The analyst, usually, defines training areas by identifying regions on the image that can be clearly matched to areas of known identity on the image. Such areas should typify spectral properties of the categories they represent, and, of course, must be homogeneous in respect to informal category to be classified. Size, shape and position must help convenient identification both on the image and on the ground. 15

Pixel located within these areas form the training sample used to guide the classification algorithm to assign specific spectral values to appropriate informational class. The advantage of supervised classification, relative to unsupervised classification, can be enumerated as follows.

First, the analyst has control of a selected menu of informational categories tailored to a specific purpose and geographic area. This quality may be vitally important if it becomes necessary to generate a classification for the specific purpose of comparison with another classification of the same area at a different date or if the classification must be compatible with those of neighbouring regions.

Second, Supervised Classification is tied to specific areas of known identity, determined through the process of selecting training areas.

Third, the analyst using Supervised Classification is not faced with the problem of matching spectral categories on the final map with the informational categories of interest. Finally the analyst may be able to detect serious errors in classification procedure; inaccurate classification of training data indicates serious problems in the classification or selection of training data, although correct classification of training data does not always indicate correct classification of other data.

The disadvantages of supervised classification are numerous. The analyst, in effect, imposes a classification structure upon the data. These defined classes may not match the natural classes that exist within the data, and therefore may not be distinct or well defined in multidimensional data space. Training data are often defined primarily with reference to informational categories and only secondarily with reference to spectral properties. A training area that is 100% asphalt may be accurate with respect to the asphalt designation but may still be very diverse with respect to shadowing, composition, density and therefore form a poor training area.

Training area selected by analyst may not be representative of conditions examined throughout the image. This may be true despite the best efforts of the analyst, especially if the area to be classified is large, complex and inaccessible. A large selection of training data can be a time consuming, even if ample resources are at hand. The analyst may experience problems in matching prospective training areas as defined on map and aerial photographs to the image to be classified.

16

Finally, Supervised Classification may not be able to recognize and represent special or unique categories not represented in the training data, possibly because they are not known to the analyst or because they occupy very small areas on the image.

5.2 Region Of Interest (ROI) and Training Data Training field are areas of known identity delineated on the digital image, usually by specifying the corner point of a square or rectangular area using line and column number within the coordinate system of the digital image. The analyst must, of course, know the correct class for each area, first assembling and studying maps and aerial photographs of the area to be classified and by investigating selected sites in the field. Specific training areas are identified for each informational category, following the guidelines outlined below.

The objective is to fix a set of pixel that accurately represents spectral variation present within each informational region. An important concern is the overall number of pixel selected for each category; as general guideline the operator should ensure that several individual training areas for each category provide a total of at least 100 pixels for each category. Sizes of training areas are important. Each must be large enough to provide accurate estimates of the properties of each informational class.

Therefore, they must as a group include enough pixels to form reliable estimates of the spectral characteristics of each class. Individual training fields should not, on the other hand, be too big, as large areas tend to include undesirable variation. To accumulate an adequate total number of training pixels, the analyst must devote more time to definition and analysis of the additional training fields. Conversely, use of large training fields increases the opportunity for inclusion of spectral inhomogeneities.

Shape of training areas are not important, provided that shape does not prohibit accurate delineation and positioning of correct outlines of regions on digital images. Usually it is easiest to define square or rectangular areas; such shapes minimize the number of vertices that must be specified, usually the most bothersone task form the analyst.

17

Location is important, as each informational category should be represented by several training areas positioned throughout the image. Training areas must be positioned in locations that help accurate and convenient transfer of their outlines from maps and aerial photographs to the digital image. As the training data are intended to represent variation within the image, they must not be clustered in favoured regions of the image that may not typify conditions registered throughout the image as a whole. It is desirable to use direct field observations in the selection of training data, but the requirement for an even distribution of training fields often conflicts with practical constraints, as it may not be practical to visit remote or inaccessible sites that may seem to form good areas for training data. “The optimum number of training areas depends upon the number of categories to be mapped, their diversity, and the resources that can be devoted to delineating training areas. Ideally, each informational category or each spectral subclass should be represented by a number of training to ensure that the spectral properties of each category are represented”8. Because informational categories are often spectrally different, it may be necessary to use several sets of training data, due to the presence of spectral subclasses. Experience indicates that is usually better to define many small training areas than to use only a few large areas.

Placement of training areas my be important also. Training areas should be placed within the image in a mode that permits convenient and accurate location with respect to distinctive features on the image. They should be distributed throughout the image so that they provide a basis for representation of the diversity present within the scene. Boundaries of training fields should be places well away from the edges of contrasting parcels so that they do not encompass edge pixels.

Perhaps, the most important property of a good training area is its uniformity or homogeneity: data within each training area should exhibit a uni modal frequency distribution for each spectral band to be used. Prospective training areas that exhibit bimodal histograms should be discarded if their boundaries cannot be adjusted to yield more uniformity. Training data provide values that estimate the means, variances, and covariances of spectral data measured in several spectral channels. For each class to be mapped, these estimates approximate the mean values for each band, the variability of each band and the interrelationships between bands.

8

M. Lillesand, Remote Sensing and image interpretation, Wiley, New York, 1979, page 264

18

5.2.1 ENVI Procedure – Region of Interests The choice to determine the regions of interest take place by attempting to show the main elements of the vegetation in urban areas. In addition to the main families of needle leaves and Broadleaves vegetation I have decided to include elements as light and dark grass which has a very specific spectral signature than the other. In addition, the components have been identified typical urban components as asphalt (roads), roofs (buildings).

To create Broadleaves ROI : 1) open Mosaic RGB 432 2) open ROI tools 3) choose new region and draw polygons on tree foliage taking care not to select shadowing elements (Figure 8) 4) analyse ROI histogram statistics about spectral homogeneity (Figure 8)

I have repeated same operations about following ROI elements: - Needle-Leaves (Figure 9) - Light Grass (Figure 10) - Dark Grass (Figure 10) - Asphalt : when identifying the ROI for all asphalt be smart to take an average of the different types of asphalt that are present in urban areas due to the different composition and age of the asphalt - Roofs: the roofs were merged from the original classification of light roofs (due to the sun) and dark tiles (shaded).

The distributions of training areas response pattern can be graphically displayed in many formats. Histogram outputs are particularly important when a spectral angle mapper is used, since it provides a visual check on the normality of the spectral response distributions. To evaluate the spectral separation between categories it is convenient to use some form of coincident spectral plots which illustrates in each spectral band the mean spectral response of each category and the variance of distribution. Such plots indicate the overlap between category respond patterns.

In order to compare ROI statistics separability and see if they are well differentiated with each other select the input image and the classes of interest. 19

A measure of the statistical separation between category response patterns can be computed for all pairs of classes and can be presented in the form of a matrix. Best quantitative expression of category information statistics is called Jeffries Matusista (JM) distance reports that must contain values from 1.9 to 2.Values below 1.9 means a conflict of values between multiple identical items. (Figure 11)

When satisfied of final result, after having selected and specified entire regions of interest (ROI) it is possible to apply spectral angle mapper technique which is described in next paragraph.

5.3 SAM (Spectral Angle Mapper) SAM is a deterministic method that looks for an exact pixel match. SAM is a physically-based spectral classification that uses an n-dimensional angle to match pixels to reference spectral. The algorithm determines the spectral similarity between two spectra by calculating the angle between the spectra and treating them as vectors in a space with dimensionality equal to the number of bands. When used on calibrated reflectance data, this technique is relatively insensitive to illumination and albedo effects. Endmember spectral used by SAM can come from ASCII files or spectral libraries, or you can extract them directly from an image (as ROI average or Z-profile spectra).

SAM compares the angle between the endmember spectrum vector and each pixel spectrum vector. Smaller angles represent closer matches to the reference spectrum. Pixels further away than the specified maximum angle threshold in radians are not classified. For example use the default threshold option single value and keep the default setting of 0.1 for Maximum Angle (radians). This parameter defines the maximum acceptable angle between the endmember spectrum vector and the pixel vector. SAM will not classify pixels with an angle larger than this value.

20

5.3.1 ENVI Procedure – SAM From SAM menu be careful to properly rename the destination file of SAM and rule image. In this case to impose rules image's threshold I used (after several attempts) different threshold values going to affect directly on rules images

1 - Select Spectral Angle Mapper from Supervised menu 2 - Choose input image 3 - Choose ROI as SLI (Spectral library) 4 - Insert multiple values classes as (for example): Broadleaves 0,4 radians Dark Grass 0.4 radians Needle Leaves 0.4 radians Light Grass 0.4 radians Roof 0.18 radians Asphalt 0.15 radians The Figure shows some Rules Image examples. (Figure 12)

21

6. POST CLASSIFICATION Classified images require post-processing to evaluate classification accuracy and to generalize classes for export to image-maps and vector GIS. Post Classification can be used to classify rule images; to calculate class statistics and confusion matrices; to apply majority or minority analysis to classification images; to clump, sieve, and combine classes; to overlay classes on an image; to calculate buffer zone images; to calculate segmentation images; and to output classes to vector layers. These procedures improve the quality of final results such as increasing the precision and separating the classes more evenly.

6.1 Confusion matrix Another area that is continuing to receive increased attention by remote sensing analyst is that of classification accuracy assessment. Historically, the ability to produce digital land covers classification or similar far exceeded the ability to quantify their accuracy. In fact, this problem sometimes precluded the application of automated land cover classification techniques even when their cost compared favourably with more traditional means of data collection. That is why a classification is not complete until its accuracy is assessed.

One of the most common means of expressing classification accuracy is the preparation of a classification error matrix (in ENVI called Confusion matrix). Error matrices compare, on a category by category basis. The relationship between known reference data (ground truth) and the corresponding results of an automated classification. Such matrices are square, with the number of rows and columns equal to the number of categories whose classification accuracy is being assessed.

The matrix stems from classifying the training set pixels and listing the known cover types used for training (as columns) versus the pixel classified into each land cover category by the classifier (as rows).9 Several other descriptive measures can be obtained from the error matrix. The overall accuracy is computed by dividing the total number of correctly classified pixels by the total number of reference pixel. Likewise, the accuracies of individual categories can be calculated by dividing 9

Envi Tutorial Confusion Matrix www.ittvis.com

22

the number of correctly classified pixels in each category by either the total number of pixels in the corresponding row or column.

Accuracy results derive from dividing the number of correctly classified pixels in each category by the number of training set pixels used for that category. User's accuracies are computed by dividing the number of correctly classified pixels in each category by the total number of pixels that were classified in that category (tow row total). This diagram is a measure of commission error and indicates the probability that a pixel classified into a given category represent that category on the ground.

It should be remembered that such procedures only indicate how well the statistics extracted from these areas can be used to categorize the same areas. If the results are good, it means nothing more than that the training areas are homogeneous, the training classes are spectrally separable, and the classification strategy being employed works well in the training areas. This aids in the training set refinement process, but it indicates little about hot to classifier performs elsewhere in a scene. One should expect training area accuracies to be overly optimistic, especially if they are derived from limited data set.

6.1.1 ENVI Procedures: how to calculate Confusion Matrix Use Confusion Matrix to show the accuracy of a classification result by comparing a classification result with ground truth information. ENVI can calculate a confusion matrix (contingency matrix) using either a ground truth image or using ground truth regions of interest (ROI). In each case, an overall accuracy, producer and user accuracies, kappa coefficient, confusion matrix, and errors of commission and omission are reported. Here, we use the ground truth ROI to assess the classification accuracy.

The ground truth ROI must be opened and associated with an image of the same size as the classification output image. The ROI are automatically loaded into the Match Classes Parameters menu. When the Match Classes Parameters menu appears, match the ground truth ROI with the classification result classes by clicking on the matching names in the two lists and clicking Add

23

Combination. The class combinations are shown in a list at the bottom of the menu. If the ground truth and classification classes have the same names, they are automatically matched.

The report shows the overall accuracy, kappa coefficient, confusion matrix, errors of commission (percentage of extra pixels in class), errors of omission (percentage of pixels left out of class), producer accuracy, and user accuracy for each class. Producer accuracy is the probability that a pixel in the classification image is put into class X given the ground truth class is X. User Accuracy is the probability that the ground truth class is X given a pixel is put into class X in the classification image. The confusion matrix output shows how each of these accuracy assessments is calculated. (Figure 13 )

Classification about vegetation condition for the year 2008 ends by this procedure (Figure 14). Difficulties of analysis stage, in spite of excellent statistical comparison, have seen noticed in distinction between Light Meadows or Grass and Broadleaves, between Dark Grass and Needle Leaves, in the presence of shadows (which determined a Non classified new class) in the presence of industrial plants which fall under light colour classes. All problems are to be solved making new classification with better values thresholds standing, checking rules image values.

6.2 Data management: Land Use overlap Having availability of data concerning year 2003 QuickBird images of Bologna's municipality and willing, in this plan, to achieve a detection change by 2008 AGEA images, we have to compare classes. This is possible having similar data with same number of classes and same kind of classes for each image. Therefore I have to effect some inclusion operations of some classes regarding 2008 classification in which we have now 6 classes (6 classes must be reduced to 4). I have to add, at first, the layer referred to agricultural land use.

The first step is to convert agricultural feature of year 2008 to same raster file (kindly provided by colleagues)as masking operations. Then is necessary build and execute new decision tree to standardize the classes. To convert agricultural features (SHP file), I converted to GRID format using spatial analyst tools thanks to ARCGIS procedures.

24

6.2.1 ArcGis Procedures 1 - Load Agricultural field shape file 2 - From spatial analyst menu click on convert features to raster selecting input shape,output path and format as Grid one 3 - Load generated raster file (Figure 15)

6.3 Decision Tree Decision tree classifiers have also been utilized to simplify classification computation and maintain classification accuracy. These classifiers are applied in a series of step, with certain classes being separated during each step in the simplest manner possible.

A decision tree is a type of multi stage classifier that can be applied to a single image or a stack of images. It is made up of a series of binary decisions that are used to determine the correct category for each pixel. The decisions can be based on any available characteristic of the dataset. For example, you may have an elevation image and two different multi spectral images collected at different times and any of those images can contribute to decisions within the same tree. No single decision in the tree performs the complete segmentation of the image into classes. Instead, each decision divides the data into one of two possible classes or groups of classes.

ENVI provides a decision tree tool designed to implement decision rules, such as the rules derived by any number of excellent statistical software packages that provide powerful and flexible decision tree generators. If the images are georeferenced, they do not need to be in the same projection or pixel size. ENVI will stack them on-the-fly. In ENVI, decision trees can be applied to multiple datasets.

25

6.3.1 Decision Tree procedures A. QuickBird 2003 classification with 4 classes: Vegetation, Urbanized, Agricultural and No classified.

B. AGEA 2008 classification with 6 classes: Needle Leaves, Light Grass (prati_chiari), Dark Grass (prati1), Broadleaves (Latifoglia), Roofs, Asphalt.

1 - From ENVI Classification menu launch Build decision tree 2 - Add children as total number of classes 3 - Use band maths expression to compute node 4 - Change name, colours and class values to each classification 5 - Select band to associate with variable 6 - From options menu execute decision tree as .IMG format. (Figure 16)

26

7. A CHANGE DETECTION CASE STUDY: RENO AND BORGO PANIGALE DISTRICTS (2003-2008) Change detection involves the use of multi temporal data sets to discriminate areas of land cover change between dates of imaging. The types of changes that might be of interest can range from short and term phenomena such as vegetation cover or urban fringe development. Ideally, change detection procedures should involve data acquired by the same or similar sensor and be reorder using the same spatial resolution, viewing geometry, spectral bands, radiometric resolution and time of day. In this case QuickBird mosaic has a pixel resolution of 2.8 meters instead AGEA mosaic has pixel resolution of 0.5meters that implies a large difference in the analysis phase especially for small parts (foliage, hanging plants, small lawns, shady corners and mixed elements that cannot be viewed with integrity).

One way to of discriminating changes between two dates of imaging is to employ post classification comparison. In this approach, two dates of imagery are independently classified and registered. Then an algorithm can be employed to determine those pixels with a change in classification between dates.

In addition, statistics and change map can been compiled to express the specific nature of the change between the dates of imagery. The accuracy of such procedures depends upon the accuracy of each classifications used in the analysis. The errors present in each of the initial classification are compounded in the change detection process.Whether image differencing or ratioing is employed, the analyst must find a meaningful change-no change threshold within the data. This can be done by compiling a histogram for the differenced or ratioed image data and noting that the change area will reside within the tails of the distribution.

A variance from the mean can them be chosen and tested empirically to determine if it represents a reasonable threshold. The threshold can also be varied interactively in most image analysis system so the analyst can obtain immediate visual feedback on the suitability of a given threshold.

27

7.1 Study Area10 Reno District (33.307 inhabitants)is the name of a south western suburb of the municipality of Bologna that extends along the right side of the river Reno, which takes its name. It consists of two former districts: Barca and Santa Viola (divided by the municipality through Viale Palmiro Togliatti), which in 1986 were combined into a single district, the Reno, exactly. The surface extension of Reno district is 5,27 Square Km) Barca's area is located in the north side from Santa Viola and bordered to the west with the municipality of Casalecchio; Santa Viola is bounded on the north by rail from Milan and by Via del Chiu. To the south and east Reno district ends at Via Don Sturzo, via Andrea Costa and the Certosa cemetery, finally surrounded to the west by Reno watercourse next to district of Casteldebole.

The right bank of the Reno river forms a border to the west and gives the area its identity. Another notable, and also problematic, element of the area is the south-west highway that separates the area, crossing it from north to south (thereby isolating one built up section near the Reno canal) and from east to west dividing the two areas. Barca is an important central focal point, with good integration between the facilities present (some of the best in the urban environment) and the Via Emilia that, while suffering from the problem of traffic, plays a central role for the Santa Viola area.

Borgo Panigale(24.935) is a district of Bologna place in the western area between the rivers Reno and Lavino with an extension of 26,166 square Km. It is the district with a lower population density .Borgo Panigale is an outlying district brimming with history and identity. Separated from the centre of the city by the large infrastructures that are its boundaries (motorway, highway, railway, the airport, the Reno river) it is, in itself, complete and self-contained and provides a variety of services that are not always to be found on the outskirts of towns. The Via Emilia main road has dictated the district’s development, and the various neighbourhoods are distributed along it: Borgo Panigale, Birra and Casteldebole, Lavino.

There is also a large high quality agricultural area south of the Via Emilia, large parts of which are owned by the Municipality, and the disused Vocational Training Centre building. Villa Pallavicini, a meeting place, land mark and integration centre for the area, is also here. With its impressive driveway it stands out in the agricultural landscape. 10

Comune di Bologna - Dipartimento Qualità della Città - Settore Ambiente, Statistiche su popolazione residente e densità per quartieri 1996-2008

28

The area of interest of this case study refers to a surface such as 9.4 squares Km that includes portions of Reno and Borgo Panigale districts. (Figure 17)

7.2 ENVI procedures: Change Detection 1 - From Post Classification menu choose change detection statistics 2 - Select Reno district 2003 classification as initial state image 3 - Select Reno district 2008 classification as final state image 4 - Add classes to be compared 5 - Select report type registered images and output path 6 - Load mask to changing analysis

29

8. CONCLUSION The classification of green areas thus obtained has the great advantage of detecting even the private garden which is currently not surveyed by the municipalities. Given the type of spectral response vegetation in relation to parts of buildings, the precision is very high.

The green areas no classified are basically of three types: green roof, foliage falling on buildings and shaded areas (small green areas in densely built)11.

These classifications, however, are not negligible compared to the total area. The extension of the methodology at Regional, at least for the larger population centers, would evaluate the extent of total and green recreational and ecological purposes.

I faced difficulties in the identification of different elements caused by small changes of land use from urban feature to vegetation feature as small lawns, hidden flowerbeds under foliage trees. The biggest changes are found from the elements with those growing urbanized, especially in a context (Reno and Borgo Panigale districts) with a low population density compared to the city center, with a series of buildings built since 2003. The different pixel resolution describes many difficulties because the changes to the QuickBird pixels are much less visible than those of Agea 2008.

The methods used in types of classification have some points of merit: the first classification with six classes was a qualitative description of the great families of trees such as deciduous and coniferous trees, the different types of grass and artificial elements as asphalt, etc. At the same time, difficulties have arisen during the class confusion matrix and accuracy of ground truth: the many class choices lead to a greater range of spectral signatures that can be mixed due to solar light, shadows, unusual items and other errors of assessment not readable by the operator during analysis of the statistics report. The qualitative classification describes a precise representation of the vegetation in the municipality of Bologna.

However, the situation of the vegetative state is positive (Figure 18), urban green compose good percentage and it is increasing over the years thanks to the protection of urban parks and a

11

Arpa Estimo e Territorio N. 7/8 - 2007 pag 41-48

30

municipal plan that provides for the maintenance of urban green such as public and private type. The figures show, with a margin of error due to the accuracy of the classification, the vegetation's percentage with a predominance of light grass and Broadleaves that equalize the total artificial elements such as asphalt roads and buildings.

The second classification used in a more quantitative method, describes the changes of land use in the period between 2003 and 2008 (Figure 19). Of course is possible to notice the transition from vegetative features to the class with a higher percentage as urbanized features but even shorter than expected (Figure 20).

I also found significant changes to the state agricultural and urban small changes state to the agricultural and urban vegetation (Figure 21, Figure 22). Here are statistical reports of change detection for its class and some photo examples that provide a representation of the work done during this training experience.

31

9. FIGURES 1 Data referring to surface extension, population and density of Bologna districts 2 True colour image example 3 False colour image example as 432 band combination 4 Spectral response of vegetation 5 Comparison of pixel resolution. Substantial differences between QuickBird (2,8m) and Agea (0,5m) 6 Boundaries on CTR sections of municipality of Bologna 7 Clipped image of entire municipality of Bologna after mosaicking procedure 8 Region of Interest. Broadleaves example, spectral graphs and photo 9 Region of Interest. Needle-Leaves example, spectral graphs and photo 10 Region of Interest. Different Grasses example, spectral graphs and photo 11 ROI separability. Values up to 1.9 indicate excellent separation 12 Rules Image example 13 Confusion matrix report 14 Classification map after SAM procedure 15 Agricultural layer to overlap on classification 16 Decision tree execution (on the left) and new classification map 17 Districts of Bologna. Areas of Borgo Panigale and Reno districts 18 Class distribution of classification supervised after SAM procedures Values are expressed in Square KM 19 Class distribution of 2003 (A) and 2008 (B) Variations of classes’ distribution with positive and negative values (C) 20 Change detection's example. From vegetation use to urbanized 21 Change detection's example. From agricultural use to urbanized 22 Change detection's example. From agricultural use or urbanized features to vegetation use

32

Superficie e densità di popolazione dei quartieri e delle zone di Bologna al 31 dicembre 2008 Quartieri Zone Borgo Panigale Navile Bolognina Corticella Lame Porto Marconi Saffi Reno Barca Santa Viola San Donato Santo Stefano Colli Galvani Murri San Vitale Irnerio San Vitale Saragozza Costa-Saragozza Malpighi Savena Mazzini S.Ruffillo Centro storico (1) Bologna (2)

Superficie (Kmq) 26,166 25,892 4,943 9,865 11,084 3,721 1,058 2,663 5,278 3,330 1,948 15,446 29,000 25,062 1,116 2,822 12,168 1,374 10,794 11,705 10,746 0,959 11,469 5,752 5,717 4,507

Popolazione residente 24.736 64.593 32.751 17.486 14.356 31.407 14.021 17.386 32.990 20.617 12.373 31.006 49.325 8.262 13.144 27.919 46.746 13.929 32.817 35.896 23.742 12.154 58.189 37.298 20.891 53.248

Densità di popolazione (Abit./Kmq) 945,3 2.494,7 6.625,7 1.772,5 1.295,2 8.440,5 13.252,4 6.528,7 6.250,5 6.191,3 6.351,6 2.007,4 1.700,9 329,7 11.777,8 9.893,3 3.841,7 10.137,6 3.040,3 3.066,7 2.209,4 12.673,6 5.073,6 6.484,4 3.654,2 11.814,5

140,845

374.944

2.662,1

(1) Costituiscono il Centro storico le zone Galvani, Irnerio, Malpighi e Marconi.

Figure 1: Data referring to surface extension, population and density of Bologna districts

33

Figure 2: True colour image example

34

Figure 3: False colour image example as 432 band combination 35

Figure 4: Spectral response of vegetation 36

Figure 5: Comparison of pixel resolution. Substantial differences between QuickBird (2,8m) and Agea (0,5m)

37

Figure 6: Boundaries on CTR sections of municipality of Bologna

38

Figure 7: Clipped image of entire municipality of Bologna after mosaicking procedure 39

Figure 8: Region of Interest. Broadleaves example, spectral graphs and photo

40

Figure 9: Region of Interest. Needle-Leaves example, spectral graphs and photo 41

Figure 10: Region of Interest. Different Grasses example, spectral graphs and photo

42

Input File: Mosaico_432_clip.img ROI Name: (Jeffries-Matusita, Transformed Divergence) Latifoglia [Green] 898 points: Prato_1 [Cyan] 1647 points: (1.79058213 1.90051833) Conifera_2 [Magenta] 135 points: (2.00000000 2.00000000) Prati_chiari [Purple] 2374 points: (1.99999990 1.99999997) coppi_all [Aquamarine] 11145 points: (2.00000000 2.00000000)

Prati_chiari [Purple] 2374 points: Latifoglia [Green] 898 points: (1.99999990 1.99999997) Prato_1 [Cyan] 1647 points: (1.99968885 1.99985197) Conifera_2 [Magenta] 135 points: (1.99999999 2.00000000) coppi_all [Aquamarine] 11145 points: (2.00000000 2.00000000)

Prato_1 [Cyan] 1647 points: Latifoglia [Green] 898 points: (1.79058213 1.90051833) Conifera_2 [Magenta] 135 points: (2.00000000 2.00000000) Prati_chiari [Purple] 2374 points: (1.99968885 1.99985197) coppi_all [Aquamarine] 11145 points: (2.00000000 2.00000000)

coppi_all [Aquamarine] 11145 points: Latifoglia [Green] 898 points: (2.00000000 2.00000000) Prato_1 [Cyan] 1647 points: (2.00000000 2.00000000) Conifera_2 [Magenta] 135 points: (2.00000000 2.00000000) Prati_chiari [Purple] 2374 points: (2.00000000 2.00000000)

Conifera_2 [Magenta] 135 points: Latifoglia [Green] 898 points: (2.00000000 2.00000000) Prato_1 [Cyan] 1647 points: (2.00000000 2.00000000) Prati_chiari [Purple] 2374 points: (1.99999999 2.00000000) coppi_all [Aquamarine] 11145 points: (2.00000000 2.00000000)

Figure 11: ROI separability. Values up to 1.9 indicate excellent separation.

43

Figure 12: Rules Image example.

44

Figure 13: Confusion matrix report

45

Figure 14: Classification map after SAM procedure

46

Figure 15: Agricultural layer to overlap on classification 47

Figure 16: Decision tree execution (on the left) and new classification map

48

Figure 17: Districts of Bologna. Areas of Borgo Panigale and Reno districts

49

Figure 18: Class distribution of classification supervised after SAM procedures. Values are expressed in Square Km

50

C) Change Detection : classes variation

B) Class. 2008

A) Class. 2003

0,3

Reno and Borgo Panigale districts

Reno and Borgo Panigale districts

0,2

veg agr

3,91

3,53

veg agr urb

urb 5,01 5,27

Square Km

0,1 0 -0,1

veg agr urb

-0,2 -0,3

0,13

-0,4

0,26

-0,5

Figure 19 : Class distribution of 2003 (A) and 2008 (B). Variations of classes’ distribution with positive and negative values (C)

51

Figure 20: Change detection's example. From vegetation use to urbanized

52

Figure 21: Change detection's example. From agricultural use to urbanized

53

Figure 22: Change detection's example. From agricultural use or urbanized features to vegetation use

54

10. REFERENCES  M.J. Canty, Image Analysis, Classification and Change Detection in Remote Sensing: With Algorithms for ENVI/IDL, Taylor and Francis, London, 2003;  J. Campbell, Introduction to Remote Sensing, Taylor and Francis, London, 2006;  M.Gomarasca, Elementi di Geomatica, Springer, New York, 2009;  M. Lillesand, Remote Sensing and image interpretation, Wiley, New York, 1979;  R. Lunetta, Remote sensing and GIS accuracy assessment, CRC Press, Boca Raton, 2004;  R. McCoy, Field methods in Remote Sensing, Guilford, New York, 2005;  R. McCoy, Resource management information systems: remote sensing GIS and modelling, Guilford, Boca Raton, 2000;  A. Spisni, Stima speditiva del verde urbano a Bologna tramite analisi di immagini QuickBird multispettrali, http://www.arpa.emr.it/dettaglio_documento.asp?id=515&idlivello=216, 2005.

55

56