The investigation of the inability to distinguish among the coniferous forest species using hyperspectral satellite image

İbreli orman türlerinin hiperspektral uydu görüntüsü ile ayırt edilebilirliğinin araştırılması

Abstract

In nowadays, the principal goal of modern forestry is to provide the capacity to be useful itself of the sustainability of forest. Consequently, current information on areas of forest should be reached in order to realize the services of forestry like the regeneration, the forest tending, and the forestation which are carried out in order to contribute to the productivity, continuity, and the protection of the forests. With recently developed technology, the remote sensing providing of the cost-effective, reliable and fast results obtains the significance information on the forest applications. Vegetation monitoring using broad band multispectral remote sensing is well established. Recently, many new sensors are being developed to increase the discrimination ability of the objects, such as the acquisition of hundreds of narrow band spectra, termed hyperspectral remote sensing or imaging spectroscopy. Imaging spectroscopy refers to data acquired by an airborne or spaceborne imaging spectrometer and the analysis techniques applied to these data in ways that exploit the instrument’s ability to resolve absorption features caused by the chemical bonds and physical structure of surface materials. In comparison to the handful of channels available with multispectral, broad band remote sensing, imaging spectrometers measure the radiation upwelling from a surface in hundreds of contiguous, narrow band width channels. The advantage offered by such spectroscopic measurements is the ability to resolve absorption features and determine their specific wavelength positions and characteristic shapes. These absorption features can be related to the material or materials causing them; thus, the materials occurring in a pixel of imaging spectroscopy data can be identified. In this study, the coniferous forest tree species were tried to be distinguished by using hyperspectral image taken by the EO-1 Hyperion at Sundiken Mountain, Turkey. The Sundiken Mountain is strongly covered with pure coniferous forests as well as mixed stands with deciduous trees. The dominant deciduous trees are oak (Quercus spp.), occurring in mixed stands with pine species stands. Anatolian Black Pine (Pinus nigra), Turkish Red Pine (Pinus brutia) and Scots Pine (Pinus sylvestris) are the dominant naturally-growing coniferous trees. Anatolian Black Pine and Turkish Red Pine occurs mixed with the oak trees. The spatial distribution of forest species on the Sundiken Mountain is changing related to the altitude and aspect. The intemperate forests at high elevations in the region receive large amounts of precipitation during the long, cold winter. At lower elevations, in Sundiken’s relatively temperate valleys, Turkish Red Pine and oak trees communities predominate. In addition to altitude, the aspect has an influence on the distribution of plants within the mountain. The atmospheric and the topographic effects in the EO-1 Hyperion image, used in this study, were corrected and converted to reflectance values using the ATCOR-4 (ATmospheric CORrection ) program. Hyperion collects data in 224 contiguous channels of approximately 10-nm bandpass over the spectral wavelength range of 0.35–2.50 μm (from visible light to near-infrared). The missing portions of the spectrum have low signal to noise due to strong atmospheric water vapour absorption, and are not used in subsequent calculations. In Sundiken, for the mean elevation of 1500 m, the Hyperion sensor measured pixels with a nominal size of 30 m at nadir view. The sensor swath width was approximately 7.5 km. Hyperion data were acquired on September 17, 2004, at approximately 10:20 a.m. local time. Our goal is to measure the ability to distinguish related to the spectral variability of coniferous types using hyperspectral data. To do this, the measurements of band depths for chlorophyll and leaf water content of each forest types obtained from hyperspectral image are examined. Finally, we classified the hyperspectral image using the spectral angle mapper (SAM) classification algorithm and compared directly to the ground truth. The Spectral Angle Mapper is a technique to classify hyperspectral data by determining the similarity between an endmember spectrum and a pixel spectrum in an n-dimensional space. Smaller angles represent closer matches to the reference spectrum. Image-based end member spectra of the main land cover types and tree species in the test area are used as input for the classification. The results obtained from the classification showed that the distinction of Turkish Red Pine from other species is satisfied while to distinguish the tree species of Anatolian Black Pine and Scots Pine are weak; because their spectral properties are very similar.

Özet

Günümüzde çağdaş ormancılığın amacı, ormanın sürekliliğini sağlayarak optimum yararlanmayı temin etmektir. Dolayısıyla ormanın verimi, sürekliliği ve korunması için yürütülen gençleştirme, bakım, ağaçlandırma gibi ormancılık faaliyetlerinin en iyi şekilde gerçekleştirilebilmesi için orman alanlarına ait bilginin güncel olması gerekmektedir. Son yıllarda teknolojinin gelişmesi ile birlikte ucuz ve hızlı sonuç alınabilinen uzaktan algılama tekniğinin ormancılık çalışmalarında önemi giderek artmaktadır. Bitki örtüsünün çok bantlı uydu görüntüleri ile izlenmesi literatürde en çok kullanılan yöntemlerden birisidir. Son zamanlarda objelerin ayırt edilebilirliklerini arttırarak daha iyi sonuçlar elde etmek amacıyla hiperspektral veya görüntü spektroskopisi olarak adlandırılan birçok yeni algılayıcı geliştirilmiştir. Objelerin spektral karakteristikleri sürekli ve daha çok spektral bantlara sahip olan hiperspektral uydu görüntüleri ile daha iyi belirlenmektedir. Bu çalışmada da Sündiken Kütlesi’ndeki ibreli orman türleri EO–1 Hyperion hiperspektral uydu görüntüsü (17.09.2004) ile ayırt edilmeye çalışılmıştır. Sündiken Kütlesi’nde baskın olarak saf ibreli orman türleri ve bunun yanında meşe ile ibreli ağaç türlerinin birlikte yetiştiği karışık orman alanları bulunmaktadır. Çalışmada yer ölçmelerine bağlı olarak Hiperspektral uydu görüntülerinden elde edilen test alanlarına ait spektral eğriler kullanılarak SAM (Spectral Angle Mapper) algoritması ile sınıflandırma işlemi yapılmıştır. Elde edilen sonuçlar Karaçam (Pinus nigra) ve Sarıçam (Pinus sylvestris) ağaç türlerinin spektral özelliklerinin çok benzer olması sebebiyle birbirinden ayırt edilebilirliğinin zayıf, Kızılçam’ın (Pinus brutia) ise ayırt edilebilirliğinin yüksek olduğunu göstermiştir.