International Geoinformatics Research and Development Journal

Rare vegetation classification of remotely sensed images, National Park,

Yelena M. G.1, Adil Y. G.2, Rustam B. R.3, Maral H. Z.4 1 R.I.S.K. Company, , Azerbaijan E-mail: [email protected] 2 SAHIL IT Company, Baku, Azerbaijan 3 Institute of Physics of the National Academy of Sciences, Baku, Azerbaijan 4 Institute of Botany of the National Academy of Sciences, Baku, Azerbaijan

Abstract

This study concentrated to develop a remote sensing and GIS method to estimate rare vegetation in Gobustan, Azerbaijan. For the study SPOT 5 of May 2007 image was procured. The paper analyses the supervised classification method Maximum Likelihood Classification algorithm. A number of steps were involved in the process of the study, including Geographical Data Base design, creation of Specialized GIS Environment and classification. Study reveals that optical remote sensing techniques could effectively be used to assess rare vegetation and map them accordingly with a reliable accuracy of 74.2%.

Keywords: Remote sensing, vegetation, classification, SPOT5

Introduction

Satellite remote sensing has become a common tool of investigation, prediction and forecast of environmental change and scenarios through the development of GIS-based models and decision- support instruments that have further enhanced and considerably supported decision making [1], [2], [3]. With the advent of new high spatial and spectral resolution satellite, new applications for precision mapping and accurate monitoring have become feasible.

Remote sensing is expected to provide us an efficient tool for vegetation monitoring. In particular, as vegetation is characterized by a mixture of vegetation, soil and water in mixed conditions, remote sensing is expected to delineate the relation between them. Rare vegetation mapping/classification is one of the most widely used applications of remote sensing. In many countries the approach has been accepted that facilitates fast and up-to-date classification of rare vegetation. Classification of land cover related to rare vegetation resource management in Azerbaijan is a priority in all aspects of vegetation mapping using remote sensing and related technology such as GIS. Additionally, information about rare vegetation distribution from satellite remote sensing has been used as the main source for further analysis in aspects of biodiversity conservation including rare vegetation rehabilitation, inventory, and catchment monitoring.

This paper describes remote sensing methodologies for rare vegetation monitoring with special emphasis on the application of SPOT5 data of vegetation classification.

Our study aims to identify and describe the extent of rare vegetation communities found within the Gobustan State National Park using GIS and Remote Sensing.. Using the accurate spatial information, our work will help to identify areas where further survey work is required and to develop mitigation strategy to reduce the impact rate of the natural and anthropogenic factors on environment.

Vol. 1, Issue 2, June 2010 Page 1 International Geoinformatics Research and Development Journal

Vegetation in GOBUSTAN

The desert communities in the Gobustan State National Park represent the most ecologically important habitat, from a botanic point of view (Figure 1). Some of vegetation within this study area now being classified as either rare or threatened and recommended for inclusion in an updated National Red List and some species are listed as globally threatened. The great age of many of the desert communities and their slow growth rate further enhance their botanic significance. The importance of this habitat type is one of the reasons that the Gobustan desert has been proposed as a State National Park, so that some level of protection is offered to this desert.

a) Tamarix b) Artemisia Lerchiana/Salsola Nodulosa

c) Suaeda Dendroides d) Alhagi pseudoalhagi

Figure 1: Rare vegetation. Gobustan

Study area

The Gobustan is located between the southern outcrops of the Mountain range and the , some 60 km south of the capital Baku as in presented in the Figure 2. The Gobustan semi- desert extends on 1780 km² (178 700 hectares) and is characterized by a semi-arid climate with continental influence and humid, cool winters and dry hot summers. The mean July temperature reaches 26.4°C and the mean January temperature 2°C in this area. Average rainfall is 200-400mm per year in Azerbaijan but can be as little as 150-200mm in semi-desert areas such as Gobustan (National Hydro-Meteorological Service 2004), [4]. The climate of this region, characterized by extreme temperatures and low rainfall, makes the land increasingly fragile with respect to anthropogenic impacts (from agricultural and industrial uses), and water management (including irrigation) has had particular impacts on the territory.

Vol. 1, Issue 2, June 2010 Page 2 International Geoinformatics Research and Development Journal

Figure 2: Study Area (Gobustan).

The Study Area at Gobustan (covering the area of 282 km2) contains a wealth of historical and archaeological sites and is also known for its rare vegetation. In 2007 Gobustan was declared a UNESCO World Heritage Site considered to be of "outstanding universal value" for the quality and density of its rock art engravings, for the substantial evidence the collection of rock art images presents for hunting, fauna, flora and lifestyles in pre-historic times and for the cultural continuity between prehistoric and mediaeval times that the site reflects.

Because of scattered distribution of vegetation fieldwork activities were done in different locations. Locations were purposefully selected to concentrate sample plots in vegetated areas. Field works carried out in the various parts of Gobustan National Park (Figure 3).

The sampling scheme was designed to collect the rare vegetation communities in the Gobustan National Park study site for combined ecological and remote sensing studies. The Field surveys were hold in accordance with preliminary data on the spreading of rare plants in the study area. Quadrates and plots assisted by satellite SPOT5 imagery have provided information on habitat types and status. Because GPS devices provided the coordinates for ground-reference data during fieldwork, the sample plots were accurately linked to SPOT5 imagery. Every plot was registered with GPS Garmin device to allow further integration with spatial data in GIS and image processing systems [5].

Based on the results of the field works the number of rare vegetation communities’ with those distributions, precise figures on population size, certain population structure have been classified (Table 1).

Vol. 1, Issue 2, June 2010 Page 3 International Geoinformatics Research and Development Journal

Figure 3: Field sample plot locations in study area

Table 1: Rare vegetation communities with habitat types.

Habitat Type Class The name of vegetation communities DESERT/SEMI-DESERT 1 Alhagi pseudoalhagi 2 Salsola Nodulosa/Artemisia Lerchiana 3 Salsola Nodulosa/Salsola Dendroides 4 Tamarix 5 Suaeda Dendroides

Geographical data base design, data automation and loading, creating specialized GIS environment

A Specialized GIS was used as software environment for performing workflow comprising of jobs connected with collecting of samples, hosting of classifier training and producing software as well as classification results analysis. Using this software, we created Geographical Data Base consisting of relevant spatial data (Orthorectified satellite multi-spectral data, ancillary data: various spectral Indexes, DEM and its derivatives as well as vector Topographical data) and Map template (Figure 4).

Vol. 1, Issue 2, June 2010 Page 4 International Geoinformatics Research and Development Journal

Figure 4: Geographical Data Base (GDB)

District boundary maps, survey maps and SPOT5 satellite images, vector polygons such as geographical areas of archaeological and historic site were graphical components of GIS Design and Application.

The Generated Land Use Map with the following layers:  Settlement;  Industrial;  Transport Infrastructure;  Greenery; etc. has been using for realization of workflows such as: - Optimization of site distribution for extraction training and test sets; - Determination of optimal spur-track (diversion route) to object of interest; - Organization of data post-processing and verification of classification results.

Classification of remotely sensed data: Classification can be considered as the process of pattern recognition or identification of the pattern associated with each pixel position in an image in terms of the characteristics of the objects or materials those are present at the corresponding point on the Earth’s surface [6].

Supervised classification of remotely sensed data: When classification is based on specific knowledge of the object features and on the decision rules in the feature space it is called supervised classification. This has been the most frequent method by which remotely sensed data of most areas has been classified [7]. Maximum Likelihood Classification algorithm was used in supervised classification. On the other hand Maximum Likelihood Classifier is the most accurate and efficient classifier. When image data meet the assumptions of parametric statistical analysis the MLC is a robust algorithm and has become a standard classifier in remote sensing data analysis [8].

Vol. 1, Issue 2, June 2010 Page 5 International Geoinformatics Research and Development Journal

Result and discussion

Qualitative evaluation of supervised classification

From the supervised classification (Figure 5) of SPOT 5 data five classes can be well identified. Resolution of SPOT5 data is 2.5 m. However, taking into account the complexities of the spectral separation between the classes have led to limit the classification to four land cover classes mentioned earlier [5]. Two of the classes were mixing with each other. After analyzing the results it would be beneficial to merge Class: Salsola Nodulosa/Artemisia Lerchiana and Class: Salsola Nodulosa/Salsola Dendroides) into one Class: Salsola Nodulosa/Artemisia Lerchiana /Salsola Dendroides .

The Image Statistical analysis of modified classification scheme shows the advances of final classification scheme and determined the best combinations of bands for separating the classes from each other [5].

Legend

Figure 5: Classification map of SPOT5 image of May 2007

Quantitative evaluation of supervised classification

Accuracy assessment Accuracy assessment is an important step in the classification process. The goal is to quantitatively determine how effectively pixels were grouped into the correct feature classes in the area under investigation. The confusion matrix is used to illustrate class agreement and error in greater detail by showing the relationship between the validation sites and the percentage of these pixels actually classified into the various classes by the maximum likelihood classifier [9], [10].

Vol. 1, Issue 2, June 2010 Page 6 International Geoinformatics Research and Development Journal

Contingency Matrix do a quick classification of the pixels in a set of training samples to see what percentage of the sample pixels are actually classified as expected [11].

A common method for classification accuracy assessment is through the use of the Contingency Matrix. During fieldworks two sample sets were collected. One training sample set was for supervised classification of image and the other ones were test sample set for accuracy assessment of classification. The Overall Accuracy for training samples is 74.2 % (Table 2).

Table 2: Contingency Matrix (training samples). Classified Data Alhagi Tamarix Suaeda SalsolaNodulosa/Artemisia pseudoalhagi Dendroides Lerchiana/SalsolaNodulosa/Salsola Dendroides Alhagi pseudoalhagi 151 0 0 28 Tamarix 0 342 0 151 Suaeda Dendroides 1 11 65 128 SalsolaNodulosa/Artemisia 5 20 11 462 Lerchiana_SalsolaNodulosa/Salsola Dendroides Column Total 157 373 76 769 Overall Accuracy = 74.2%

The accuracy in this classification suggested that this strategy for the selection of training samples, modification of classification scheme used were importance to perform better classification result.

Conclusion

If the findings discussed in earlier sections are to be credible, they should be shown to produce acceptable results for independent test datasets. The validation dataset consisted of randomly-selected samples. The number of samples was determinate on the basis of experience. Overall Accuracy for test samples was found of 70.9%. From the study it is revealed that desert vegetation in the Gobustan National Park can be assessed and mapped using SPOT5 image. For better result with higher accuracy other sensors with higher spectral and spatial resolution can be studied. Using the accurate spatial information, our work will help to identify areas where further survey work is required and to develop mitigation strategy to reduce the impact rate of the natural and anthropogenic factors on environment.

Acknowledgements

This work was supported by the Planet Action and the Idea Wild non-profit organizations for their support by donating satellite images, GIS software and equipment, which provided recourses for the research that led to this paper.

References

[1] Ayad Y., Remote Sensing and GIS in modeling visual landscape change: a case study of the north- western arid coast of Egypt, Landscape and urban planning, 73 (4), 307, (2005).

[2] Douglas C., History and Status of Aerial and Satellite Remote Sensing and GIS in the Inventory and Evaluation of Cultural Sites, Proceedings of the 8th Annual US/Icomos International Symposium Heritage Interpretation, 356, (2005).

[3] Hadjimitsis D.G., Clayton C.R., and Retalis A.G., Darkest pixel atmospheric correction algorithm: a revised procedure for environmental applications of satellite remotely sensed imagery, Proceedings 10th International Symposium on Remote Sensing, Barcelona -SPAIN, organized by NASA, SPIE CONFERENCE, 414, (2003).

[4] National Hydro-Meteorological Service: Climatic Data for Kürdamir and Baku Climatic Station (unpublished data, personally obtained from National Hydrometeorology Service Department at the Ministry of Ecology and Natural Resources), Baku, Azerbaijan, (2), (2004).

Vol. 1, Issue 2, June 2010 Page 7 International Geoinformatics Research and Development Journal

[5] Gambarova, Y.M., Gambarov, A.Y, Rustamov, R.B., and Zeynalova, M. H. Remote Sensing and GIS as Advance Space Technologies for Rare Vegetation Monitoring in Gobustan State National Park, Azerbaijan, Journal of Geographic Information System (JGIS), 93-99, (2010).

[6] Mather, P. M. Computer Processing of remotely Sensed Images-An Introduction, John Willy &Sons, Toronto, 167-173, (1999).

[7] Green, E. P., Clark, C.D., Mumby, P.J., Edwards, A.J and Ellis, A. C. Remote Sensing Technique for Mangrove mapping, International Journal of Remote Sensing, 5(19), 935-956, (1998).

[8] Treitz, P.M., Howarth, P.J, Rotunno, O. and Soulis, E.D. Agricultural crop classification using SAR tone and texture statistics. Canadian Journal of Remote Sensing 26(1), 18–29, (2000).

[9] Jensen, J.R. Introductory Digital Image Processing: A remote sensing perspective 3rd Prentice Hall New Jersey, (2005).

[10] Lillesand, T. M. and Kiefer, R.W. Remote sensing and image interpretation. (4th) New York: John Wiley & Sons, (2000).

[11] Smits, P. C., Dellepiane, S. G., and Schowengerdt, R. A. Quality assessment of image classification algorithms for land-cover mapping: A review and a proposal for a cost-based approach, International Journal of Remote Sensing, (20), 1461-1486, (1999).

Vol. 1, Issue 2, June 2010 Page 8