USING LANDSAT MULTISPECTRAL IMAGES IN ANALYSING FOREST VEGETATION

Transcription

1 Technical Sciences 243 USING LANDSAT MULTISPECTRAL IMAGES IN ANALYSING FOREST VEGETATION Teodor TODERA Traian CR CEA Alina NEGOESCU Faculty of Geography of Tourism, Sibiu, Romania ABSTRACT The article contains certain aspects regarding the use of multispectral images in analyzing the forest vegetation from Gura Râului area, Sibiu County. The paper presents the basic principles of processing multispectral images, how to conduct analysis of vegetation types and some aspects related to multispectral image georeferencing. In the case study, we have presented a quantitative analysis of the forest vegetation in the studied area by the method of thematic classification of informational content of multispectral images. The case study also includes a qualitative analysis of forest vegetation. In order to carry out the qualitative analysis we have used NDVI, RVI and PVI spectral Difference Vegetation Indices, on the basis of which we have highlighted the quality levels of the analyzed vegetation. Based on the results of thematic classification we have carried out an analysis of multispectral images from 2000 and 2007, obtaining conclusive results regarding the forest area studied. We have also described the process of qualitative classification of informational content of multispectral images based on Difference Vegetation Indices. KEYWORDS: multispectral image, georeferencing, thematic classification, qualitative classification 1. Introduction The article contains certain aspects concerning the use of LANDSAT multispectral satellite images in analyzing forest areas and identifying species of forest vegetation and also the variations over time of areas occupied by them, as well as the qualitative and quantitative analysis of forest vegetation.

2 244 Technical Sciences Since multispectral satellite images are obtained based on the spectral reflectance receipt of forest vegetation, in many regions of the electromagnetic spectrum (in several spectral bands), they can be used for the automatic identification of categories and quality of vegetation in the analyzed area. It is known that the spectral reflectance is the ratio between the radiation flux reflected by the surface of the investigated object (researched forest area) and the radiation flux incident on the surface of the investigated object, ratio that can take values between 0 and 1. The spectral reflectance of forest vegetation coverage varies depending on the type of vegetation analyzed (depending on the species of trees that compose the analyzed forest area) and makes it possible to identify areas covered with forest vegetation of different species [1]. The methods used for identifying classes of forest vegetation and for quantitative and qualitative analysis of this category of vegetation are those based on thematic classification, which aims at developing themed images where each pixel is assigned (based on spectral response) to a particular class of objects [2], [3]. Researching forest vegetation using multispectral satellite images requires several spectral channels in which the investigated area is recorded. 2. Basic Principles Multispectral image recordings (LANDSAT, SPOT, etc.) is characterized by a series of advantages, such as: include large surfaces of land making it possible to study the overall features of the recorded territory; allow imaging for hard to reach areas, inaccessible or hostile to man; can be obtained (depending on sensor) at any time of year or day; can reveal novel features of objects or phenomena being studied; allow notification of rapidly evolving phenomena in time; can be digitally recorded and automatically interpreted with a specialized software obtaining real-time images required on the researched land. Analysis of forest vegetation based on multispectral satellite imaging is an ultramodern, efficient and accurate method of processing content information of these images, which allows complex and detailed qualitative and quantitative analyses regarding classification of species [4], [5], [6]. Generally, in order to carry out quantitative analysis of forest vegetation we use supervised (operator assisted) and unsupervised (unassisted by operator) classification that allows a thematic classification of forest area by species of trees (fir, beech, oak, etc.). However, an analysis is performed to increase or decrease forest areas in time. For qualitative analysis of forest vegetation we use the Normalized Difference Vegetation Indices, mathematically determined by arithmetic operation between certain spectral bands on which we can identify (extracts), from the content of multispectral images, qualitative elements of vegetation surface, by comparing and highlighting the spectral signature of these elements in the near infrared spectral region (NIR) and red spectral region (R), areas where the behavior of these elements is different [7], [8]. The case study, presented in this article, includes Gura Râului land area, located in the trapeze L D, Sibiu County. For the multispectral analysis of the forest vegetation in the researched area we used Landsat multispectral images recorded in the years 2000 and 2007, in the seven standard spectral bands, with 1 G preliminary processing level (radiometrically and geometrically corrected). For the processing of multispectral images we used LEOWorks software to classify and identify forest categories and IDRISI software (Software created by Clark University, USA) for qualitative analysis operations on vegetation based on vegetation indices.

3 Technical Sciences Analysis of Forest Vegetation Types Analysis of forest vegetation types can be made through thematic classification of informational content of multispectral satellite images with the help of specialized software. The process of analysis of vegetation types involves performing georeferencing the multispectral images used and then caring out the quantitative thematic analysis of the content of georeferenced images by a supervised and an unsupervised classification of images from the red and near infrared spectral bands of the electromagnetic spectrum Georeferencing Multispectral Images First operation that needs to be performed in processing multispectral images is georeferencing these images, process by which the multispectral image is brought to the map scale at which the analysis is done [9]. According to the software used to perform georeferencing we can select either relations of polynomial transformation, relations of Helmert transformation or relations of affine transformation. In the case of multispectral Landsat images we have selected and used the relations of polynomial transformation [10], [11], [12]. First we have selected the cartographic projection used in georeferencing (Figure no. 1. a) and we have introduced its parameters (it has been chosen the stereographic projection of the year 1970), then we have performed image georeferencing based on 4 reference points for each image (Figure no. 1. b) with topographically determined coordinates in the geodetic system of coordinates of the selected cartographic projection. Figure no. 1. a. Image georeferencing choosing the cartographic projection

4 246 Technical Sciences Figure no. 1.b Image georeferencing introducing reference point coordinates 3.2. Identification of Vegetation Type Through Supervised Classification In supervised classification (operator assisted) the forest vegetation classes that are searched in the informational content of images are known in advance on certain restricted areas in the image (named test areas or sites). In other words, the user identifies (by polygon selection) several known areas on image that are characteristic to each class of established details. Through image analysis each pixel in the image is classified in one of these classes [13], [14]. For the supervised classification, in this study we used Landsat images from the years 2000 and 2007, band 3 specific to red (R) visible spectrum and band 4 specific to near infrared (NIR) spectrum. Supervised classification of digital images requires the following operations: feature selection, which involves selecting the useful information (of the vegetation area) used to classify the image content; selection of classification type, consisting in decomposing the space of features in disjoint subspaces so that any pixel to belong to one of the classes. There are three classification types: geometric classification based on measuring the distance between unknown pixel and a median vector, parallelepiped classification based on the probability that a pixel belongs to a certain class. Following the supervised classification we have identified and extracted 3 types of predominant classes (spruce, beech and hornbeam, oak). We noticed that out of the total area with forest vegetation 18.79% is occupied by spruce, 10.92% by beech and hornbeam and 14.25% by oak (Figure no. 2). Figure no. 2 Results of supervised classification (obtained from LANDSAT multispectral images in 2007, band 3 red and band 4 near infrared)

5 Technical Sciences Identification of Vegetation Type Through Unsupervised Classification Unsupervised classification (unassisted by operator) of multispectral images involves the creation of groups of pixels that represent geographic features, without knowing a priori what is being classified. Practically, pixels are organized into class clusters and then are grouped into clusters [15]. The result of the unsupervised classification, from images in band 3 specific to visible red (R) spectrum and band 4 particular to near red (NIR) spectrum, is a thematic image organized in classes of forest areas (Figure no. 3). Figure no. 3 Result of unsupervised classification (LANDSAT multispectral images from the year 2007, band 3 red and band 4 near infrared) Following the unsupervised classification (unassisted by operator) of images we have identified 5 types of predominant classes (fir, beech, oak, beech and oak, shrubs). We have found that out of the total area with forest vegetation of the image 25.86% is occupied by fir, 7.25% by beech, 14.75% by oak, 5.96% by beech and oak and 13.96% by shrubs. 4. Evolution of Forest Areas The evolution of the forest areas in the case study conducted was highlighted through the comparative analysis of the results of multispectral image processing from the years 2000 and 2007, [16]. For this purpose we have made the difference between the results of the unsupervised classification from the years 2007 and 2000, of the spectral bands 3 (R) and 4 (NIR) from the two years [17]. Following this temporal analysis of multispectral images, small changes were highlighted in the expansion or reduction of forest areas in the studied area during the period These changes in percentages are: 1.07 % expansion of area covered with fir;

6 248 Technical Sciences 1.62 % reduction of areas covered with beech; 1.91 % expansion of area covered with oak; 0.31 % reduction of areas covered with beech and oak; 1.16 % reduction of areas covered with shrubs. The most significant changes occurred in the non-forest vegetation, thus: % expansion of area covered with grasslands; 3.47 % reduction of areas covered with crops; 6.45 % reduction of areas covered with meadows. 5. Qualitative Analysis of Forest vegetation For the qualitative analysis of forest vegetation we used NDVI, RVI and PVI Differentiation Indices, applied in processing images from the red (R) spectral band and near red (NIR) spectral band [18], [19] Qualitative Analysis of Forest Vegetation Using NDVI NDVI (Normalized Difference Vegetation Index) used in the qualitative analysis of forest vegetation is a normalized spectral index of vegetation that can separate vegetation from the soil not covered by vegetation (Fig. 4). Figure no. 4 Vegetation classification based on NDVI Following the information classification from the multispectral images, six quality levels of vegetation resulted: healthy, moderated (medium density), rare (lack of water), reduced (dry or affected by diseases), without vegetation (plowing, etc.) and aquatic surfaces Qualitative Analysis of Forest Vegetation Using PVI PVI (Perpendicular Vegetation Index) is an index based on the distance at which the image pixels are located from the soil line (the soil line is a description of the spectral signature of soil, in red and

7 Technical Sciences 249 infrared, within the rectangular system of coordinates) obtained by linear regression between near infrared (NIR) band and red (R) band of an area known as being without vegetation. The pixels located close to the soil line are considered as being soil, and those located at distance from the soil line are considered vegetation [20], [21]. The result of vegetation classification based on such an index is presented in figure no. 5. Figure no. 5 Vegetation classification based on PVI 6. Conclusions Performing a quality spectral analysis of the informational content of multispectral images requires knowledge of their features and resolution. In this respect, the images contrast is the characteristic that influences the identification of topographic details of the recorded Earth s surface, and the image resolution is the one that can influence the accuracy of planimetric determination of areas. Accuracy of georeferencing images depends on the accuracy of determining the reference points and identifying them on the image. The coordinates of these reference points can be extracted from large scale maps or topographical plans. It is recommended that these reference points to be determined in the field by topographic methods with modern equipment. The choice of the method of classification for the quantitative analysis of multispectral images should be done according to categories of vegetation that are intended to be identified. The choice of differentiation vegetation indices should be

8 250 Technical Sciences made based on the type of vegetation to which they are sensitive and the quality of the multispectral images used. An appropriate choice of these spectral indices will lead to conclusive results, which will result in a sensitive highlighting of the differentiation of the quality levels of analyzed vegetation. REFERENCES 1. Teodor Todera and Vasile Dragomir, Teledetec ie i fotointerpretare, (Sibiu: Lucian Blaga University Publishing House, 2002). 2. Constantin Vertan, Prelucrarea i analiza imaginilor (Bucharest: Universitatea din Bucure ti Publishing House, 1999). 3. Aurel Vlaicu, Prelucrarea digital a imaginilor, (Cluj-Napoca: Albastr Publishing House, 1997). 4. Valentin Donis, Procesarea numeric a imaginilor, (Ia i: Azimuth Publishing House, 2004). 5. John R. Jensen, Introductory Digital Image Processing, (New Jersey: Prentice Hall, 2005). 6. Thomas M. Lillesand, Ralph W. Kiefer and Jonathan W. Chipman, Remote Sensing and Image Interpretation, (New York: John Wiley, 2004). 7. John R. Jensen, cit. ed. 8. Thomas M. Lillesand, Ralph W. Kiefer and Jonathan W. Chipman, cit. ed. 9. Teodor Todera, Vasile Dragomir, cit. ed. 10. Alexandru Badea, Teledetec ie, (Bucharest: USAMV Publishing House, 2011). 11. Andrei Mihai Bogdan, Teledetec ie. Introducere în procesarea digital a imaginilor, (Bucharest: Universitatea din Bucure ti Publishing House, 2007). 12. Valentin Donis, cit. ed. 13. John R. Jensen cit. ed. 14. Thomas M. Lillesand, Ralph W. Kiefer and Jonathan W. Chipman, cit. ed. 15. Ibidem. 16. Valentin Donis, cit. ed. 17. Ibidem. 18. Ibidem. 19. John R. Jensen, cit. ed. 20. Ibidem. 21. Thomas M. Lillesand, Ralph W. Kiefer and Jonathan W. Chipman, cit. ed.