DIRECTIONS

Download Data

To obtain the subset of the original data (image and documention files) that contains the windowed area necessary to complete this exercise in IDRISI for Windows 1.0 (nitwin.zip) click here. Approximately 20 megabytes of free computer space is needed to download and unzip these files.

Explanation on how to use zip files

To download a copy of unzip for use on a UNIX, DOS or MacIntosh platform go to the ftp site located at ftp.uu.net:/pub/archiving/zip.

FTP SITE

Idrisi Implementation Notes

DATA CHARACTERISTICS

This exercise deals with a portion of a full Landsat image. This portion of the image covers approximately 7,760 square kilometers of northern Italy. Specifications of the raw Landsat data are given below.

Original Landsat Image:

Landsat Thematic Mapper - Landsat 5

This exercise employs a windowed area of the Landsat Thematic Mapper scene, extending from row 3504 to row 5748 and column 2556 to column 6396.

The Landsat data used in this project were acquired by the European Space Agency, at Fucino, Italy, under license from EOSAT Corporation. This image presents conditions over northern Italy at approximately 10:00 a.m. on April 7, 1994. The Landsat thematic mapper provides seven spectral images covering the scene. Six of these spectral bands (bands 1,2,3,4,5, and 7) represent solar energy reflected from the Earth's surface, and the seventh (number 6) represents energy emitted from the Earth's surface as a function of surface temperature. Each spectral band file represents 2245 rows of pixels containing 3841 columns of data. These files of raw data are each over 8.6 megabytes in size. Thus all seven raw data files require almost 61 megabytes of free computer disk space. Since these data will be used to generate new data and image maps, it is apparent that one will need a fairly large amount of free disk space to work with this total scene.

If the free disk space is a problem, you can download and use a subset of the original data. These subset images are "windows" from the original scenes (from column 1119 row 307 to column 2601 row 1621). Each band is about 1.95 megabytes in size.

The three thematic mapper spectral bands from the visible portion of the spectrum (450-520, 520-600, and 630-690 nm) and the three near-visible infrared spectral bands (760-900, 1550-1750, and 2080-2350 nm) all represent reflected solar radiant electromagnetic energy, and have a surface resolution of 30 meters. The resolution of the thermal spectral band (10,400 - 12,500 nm) is 120 meters. The spectral windows for each band number designation is given below.

• Landsat band 1 = 450 - 520 nanometers
• Landsat band 2 = 520 - 600 nanometers
• Landsat band 3 = 630 - 690 nanometers
• Landsat band 4 = 760 - 900 nanometers
• Landsat band 5 = 1550 - 1750 nanometers
• Landsat band 6 = 10,400 - 12,500 nanometers
• Landsat band 7 = 2080 - 2350 nanometers

THE LANDSAT SCENE AT A GLANCE

OBSERVE THE TOTAL LANDSAT SCENE:

1. Use any image processing software which is capable of reading the digital image files, and displaying a decimated version of the raw image on the computer monitor. Observe each of the seven spectral band images, and note the following features to become familiar with the area covered by this Landsat scene. You may wish to display the raw data of each of the seven subscenes, but we will soon improve the appearance of these "pictures". Idrisi Implementation Note #1

For the purposes of this exercise, the spectral bands of this subscene have been named NIT# (standing for "Northern Italy", with the number referring to the Landsat Thematic Mapper Spectral Band). If you are using the subset data, the spectral bands are named MILNIT# (standing for the city of Milan located in the center of the window).

PROJECT OBJECTIVES:

Using the Landsat subscene prepared above, we will do the following:

1. Prepare the "images" for better visual interpretation.
2. Classify the imaged area to distinguish urban land use from other land use and land cover surfaces.
3. Prepare an estimate of the total human population living in the imaged area.
4. Recognize the limitations and possible causes for analysis errors.

You may wish to display small portions of the imaged subscene at a higher resolution to get a feel for the appearance of the area. However, the quality of the imagery will be much better after the next step in processing the data.

Objective 1: Prepare the "images" for better visual interpretation.

2. Prepare a histogram for each of the seven spectral bands of the subscene and note the minimum and maximum digital values for each of the spectral bands. [If possible, also notice the DN value which has only 5 percent of the pixels with lower values, and a second DN value which has only 5 percent of the pixels with higher values.] Once these minimum and maximum digital values have been determined for each spectral band , perform a stretch for each band. It is suggested that you try either the absolute minimum and maximum values of the file to stretch ("linear stretch"), or enter the lower 5 percent figure and the upper 95 percent figure ("linear with 5% saturation stretch").Idrisi Implementation Note #2

The stretch will increase the visual contrast of the image and provide a file which the your computer can display with greater clarity. Notice that each new stretched images will also occupy a great deal of computer disk space. You may wish to observe these images and then delete some of them. It is suggested that you keep only the band 4 stretched file, since this may be useful for general reference.

[The following are all stretched images.]

Band 1(nit1st)

Band 2(nit2st)

Band 3(nit3st)

Band 4(nit4st)

Band 5(nit5st)

Band 6(nit6st)

Band 7(nit7st)

Objective 2: Classify the imaged area to distinguish urban land use from other land use and land cover surfaces.

There are numerous ways to classify a digital image. Some of these computer algorithms include the parallelpiped, minimum distance to the mean, and the maximum likelihood classification techniques. One can attempt both supervised and unsupervised classification techniques depending upon the software and understanding available to the operator. You are free to experiment as you wish. The author of this exercise has had some success by preparing training sites over band 4, and using a maximum likelihood supervised classification. More specific instructions concerning these procedures are given below.

Classified Image (nitclass)
[Hint: Prior to actually choosing training sites, it will be very helpful to observe agricultural areas in a color composite mode (see below). Vegetated cropland will be obvious when contrasted with the cropland which has no apparent surface vegetation in early April. These latter fields would be classified as barren land.]

Identification of training sites:

3. First, window to various areas of spectral band 4 to find regions which are typical of the following land cover types.

• Class 1 = Urban
• Class 2 = Cropland (vegetated)
• Class 3 = Water
• Class 4 = Barren (plowed field without crops)
• Class 5 = Forest

Using the procedures of your chosen image processing software, identify training sites for each of the five land cover classes listed above. Perhaps 3 training sites for each of the five classes would be appropriate. Many image processing software packages will allow the identification of training sites while a color image is displayed on the computer monitor. If you have this capability it is suggested that you use an RGB=4,3,2 or an RGB=7,4,3 color composite. If your software displays a single monochrome image when choosing training sites, it is suggested that you use the band 4 image as a backdrop for the delineation of training sites. When estimating the human population of the region, in objective 3 below, you will be most concerned with accurately classifying the urban land cover of cities and villages.Idrisi Implementation Note #3

False Color Composite (RGB = Bands 7, 4, & 3)(nit743)

False Color Composite (RGB = Bands 4, 3, & 2)(nit432)
Classification of the subscene: As a first approximation, the author used a maximum likelihood classification. The resultant classification assigned the pixels of the subscene in the following way. Please notice that if you are using the smaller subset images these numbers will be different.

• Class 1 = Urban = 26.1 % of the subscene
• Class 2 = Cropland = 32.9 % of the subscene
• Class 3 = Water = 1.2 % of the subscene
• Class 4 = Barren = 34.9 % of the subscene
• Class 5 = Forest = 4.9 % of the subscene

Your classification may be fine tuned with varying alpha levels, or by changing the number of standard deviations used in parallelpiped or minimum distance to the means classification algorithms. It will be necessary to look at your classified image in various windows of the total subscene and compare these classifications with your interpretation of color composite images, or monochrome images of the same windowed areas. When you are satisfied with your classification, be sure to record the number of pixels assigned in each class.

Objective 3: Prepare an estimate of the total human population living in the imaged area.

To estimate the human population living in the Milan subscene area, you will first determine the population density of the urban landscape. To do this determine the area of the city of Milan located near the center of the Milan subscene.

4. Display a windowed picture on the computer monitor which clearly distinguishes Milan from its surrounding rural landscape. Measure the diameter of the city of Milan in the following way. It will necessary to measure several transects across the city of Milan. Use the utilities of the digital image processing software to obtain a read-out of the pixel column and row at each end of at least four straight lines which transect the city of Milan. Thus, measure four transects in the following directions: north-south, east-west, northwest-southeast, northeast- southwest. These measured distances will be averaged to determine the average diameter of the city of Milan. If you are uncomfortable with only 4 sample diameter transects, you can add additional transects. Idrisi Implementation Note #4

The following discussion explains one method of determining distance on a remotely sensed image. This method involves recording the column and row numbers of two separate points, and calculating the distance between the points. The column numbers represent the X coordinates, and the row numbers represent the Y coordinates.

Record the X and Y coordinates for the ends of each linear segment of distance between two points on the image.

For each linear distance on the image, calculate its length, in meters, using the following formula.

SQRT ((X1-X2)(X1-X2) + (Y1-Y2)(Y1-Y2))

The result of this formula will yield the distance between the two points in units of image resolution. Since the pixel resolution of these Landsat thematic mapper images is 30 meters, multiply this answer by 30 to obtain a final answer in meters. Change this to kilometers by dividing by 1,000.

Average radius of Milan = ____________________________ km

Calculate the average of the Milan diameter transects, and assume this to be the average diameter of the "circular" city. The area of a circle is expressed as pi times the square of the radius. Pi equals 3.141592654. The radius of Milan would be half the calculated average diameter. Multiplying the square of the radius (in kilometers) times pi will yield the area of Milan in square kilometers.

Area of Milan = ____________________________ Square km

The approximate population of Milan, Italy at the time of this image acquisition was 1,371,008 people. Determine the population density of Milan by dividing the population figure by the total area of the city.

Population density of Milan = ______________________ people/Square km

As a general assumption, we will assume that most people living in the imaged subscene area occupy dwellings in the cities, towns, or numerous small agricultural villages which are distributed across the landscape. You have previously determined the approximate number of pixels characterized by "urban" landscape. Each 30 meter pixel represents 900 square meters of ground area. This is 0.0009 square kilometers. Multiply the number of classified urban pixels by 0.0009 to determine the total area of urban landscape. [Of course there are inherent errors involved with this method. Primarily many paved surfaces of roads, etc will be classed as urban. Also, village parks and neighborhoods with lots of trees and biomass may not be classified as urban. In general, these inherent errors will usually cancel each other. More discussion of accuracy and errors will follow at the end of this exercise.]

Total area of "urban" land cover in the Northern Italy subscene = _____________Square km

Since you have already calculated the number of people expected in each "urban" square kilometer of the subscene (i.e., the population density of Milan), calculate the total human population of the Milan subscene.

Total human population in the Northern Italy subscene = __________________ people

Objective 4: Recognize the limitations and possible causes for analysis errors.

Northern Italy Subscene: It is obvious that the accuracy of this approach of population estimation is highly dependent upon the accuracy of the image classification. Review your process of image classification, and consider possible alternate classification procedures (or just additional classes) which might improve the final results of this analysis. Prepare a statement suggesting possible ways to improve the results of this population estimate procedure.

Discussion

This landsat image does have characteristics which make it more difficult to digitally interpret than many remotely sensed images. These characteristics include:

1. There are many different environmental situations displaying a variety of surface reflectance (and emittance in the case of band 6) characteristics.
2. The image date of April 7 displays agricultural regions which have both vegetated surfaces and barren fields. in the mid summer all of the agricultural terrain is generally covered with vegetation, and there is more obvious contrast between the rural agricultural country side and the adjacent urban landscape.
3. Flooded fields, moist soil, and cloud shadows in the southwestern portion of the imaged area make the interpretation and digital classification if the scene more difficult.

Suggest at least three additional image interpretational projects which would involve the data provided by this Landsat image. [For instance, locate the many airports from maps of the region. Locate and measure the maximum runway length of each of these airports from the imaged data. (This was my idea, you can think of three additional ones.)]

Support Resources:

The following provide background information and orientation for students and faculty working on this exercise. Road maps, such as item number one below, can be found at better book shops. The material available through the internet is continuously expanding. Students can search and find a wealth of information using the World Wide Web.

1. Kummerly & Frey, 1:500,000 road map of Northern Italy, ISBN 3-259-01118-8, Bern, Switzerland

2. 2. Use your internet browser to find information on the city of Milan and the Italian regions of Lombardia. These internet World Wide Web sites will provide access to maps and general information about this region of Italy. Start with the region of Lombardia, which includes Milan. Then also investigate Piemonte, Valle D'Aosta, Trentino-Alto Adige, Veneto, and Emilia-Romagna.

Students may also wish to search the Web for information on Switzerland and the Cantons of Switzerland.

References:

Campbell, James, 1987 Introduction to Remote Sensing, Guilford Press.

Eidetic Digital Imaging Ltd. 1993. RSVGA Image Processing Software, Eidetic Digital Imaging Ltd: Brentwood Bay, British Columbia, Canada.

Jensen, John R., 1996, Introductory Digital Image Processing; a Remote Sensing Perspective, Prentice-Hall.

Lillesand, T. and Keifer, R. 1994, Remote Sensing and Image Interpretation, second edition, John Wiley & Sons.

Answers to Questions:

Average radius of Milan = 19.475 km

Area of Milan = 1191.5 Square km

Population density of Milan = 1150.657 people/Square km

Total area of "urban" land cover in the Northern Italy subscene = 2022 Square km

Total human Population in the Northern Italy subscene = 2,326,651 people
Prepare a statement suggesting possible ways to improve the results of this population estimate procedure.

1. Add a land use classification for flooded fields, and perhaps another for areas of cloud shadow.
2. Mask areas to the southwest of the Po River which seem to add confusion to the classification process by identifying dark wet fields as "urban".
3. Perhaps a more accurate assessment of the area of some of the medium sized urban centers, and a comparison to their published populations statistics would be more appropriate than basing the total analysis upon the characteristic of just one first order city.

Suggest at least three additional image interpretational projects which would involve the data provided by this Landsat Image.

1. Rice is often grown in the southwestern portion of this image. Many of these rice fields are flooded with turbid water at the time of this image. Attempt to estimate the total area of this image (in square kilometers) which is devoted to rice cultivation.
2. The fairways of golf courses often display a distinctive pattern of linear strips of well irrigated and manicured grass. Attempt to locate several golf courses present in this image.
3. Identify examples of first, second, third, and fourth order cities within the imaged area. Does the distance between cities of the same order increase or decrease as the order number increases? Identify examples to support your answer, and explain why this is the case.