EKSTRAKSI INFORMASI PENUTUP LAHAN AREA LUAS DENGAN METODE EXPERT KNOWLEDGE OBJECT-BASED IMAGE ANALYSIS (OBIA) PADA CITRA LANDSAT 8 OLI PULAU KALIMANTAN
<p class="abstrak">Ekstraksi informasi penutup/penggunaan lahan area luas seperti di Pulau Kalimantan umumnya terkendala oleh variasi nilai spektral di beberapa area yang berbeda, serta sulitnya mendapatkan hasil perekaman yang bebas dari awan. Klasifikasi visual, meski memberikan hasil yang baik, merupakan pekerjaan yang membutuhkan waktu dan tenaga yang relatif banyak, belum lagi potensi pengaruh subjektifitas interpreter. OBIA yang sudah mulai diterima dan banyak digunakan dalam klasifikasi digital bisa menjadi alternatif tambahan selain interpretasi visual maupun analisis digital berbasis piksel konvensional. Penelitian ini menggunakan data Landsat 8 OLI <em>orthorectified </em>yang telah melalui proses <em>mosaicking</em> dan <em>cloud masking</em> untuk mendapatkan citra satu Pulau Kalimantan yang bebas awan. <em>Layer </em>NDVI, MNDWI, NDBI, BSI, SAVI, dan <em>Built-up Index</em> kemudian diturunkan dari data Citra Landsat untuk dimasukkan ke dalam tahap segmentasi dan klasifikasi. Segmentasi dilakukan dengan menggunakan algoritma <em>Multiresolution Segmentation</em> dan <em>Spectral Difference Segmentation</em>. Klasifikasi dilakukan dengan menggunakan serangkaian <em>multilevel threshold</em> yang disusun dalam bentuk <em>decision tree</em>. Empat belas kelas penutup/penggunaan lahan kemudian berhasil diekstrak, dengan nilai <em>overall accuracy</em> 77,65%. Metode yang digunakan juga menunjukkan akurasi yang tinggi untuk kelas hutan lahan kering, perkebunan, kebun campur dan semak belukar dengan nilai akurasi di atas 80%. Hasil ini menunjukkan bahwa metode ini bisa dijadikan sebagai alternatif dalam mengidentifikasi dan mengekstrak informasi tutupan vegetasi untuk kegiatan pemetaan area luas.</p><p class="katakunci"><strong>Kata kunci: </strong>OBIA, area luas, perubahan penutupan/penggunaan lahan, citra landsat, <em>decision tree</em></p><p class="judulABS"><strong><em>ABSTRACT</em></strong></p><p class="Abstrakeng"><em>Large area landuse/landcover extraction such as on the island of Borneo using Landsat 8 data are generally constrained by the great variations in the spectral values, due to the vast use of different scenes with different acquisition time, as well as the fact that it almost impossible to get a completely cloud-free image of the whole island. Visual classification, despite the good results, is a labour-intensif job that requires a huge time and effort, not to mention the potential influence of interpreter’s subjectivity. While the pixel based digital classification suffer from“salt pepper” effect as well as almost exclusively relied on spectral information, OBIA has been accepted and widely used in digital classification as an alternative for the visual interpretation and conventional pixel-based classification, with its ability to use additional contextual information. This study aimed to used OBIA method on Landsat 8 OLI cloudfree mosaic dataset for the whole Borneo region to create a landuse/landcover map using both spectral and contextual information, as well as ancilarry DEM data. Additional layers of NDVI, MNDWI, NDBI, BSI, SAVI, and Built-up Index were then derived from Landsat data to be used in the segmentation and classification process. Multiresolution Segmentation algorithm and Spectral Difference Segmentation were then conducted respectively. The classification wasdone by using a series of multilevel crisp classification using thresholds in the form of a decision tree. Fourteen of landuse/landcover classes were then successfully extracted, with a value of 77.65% on overall accuracy. The proposed method showed reasonable high accuracy for the forest, plantation, mixed garden and shrub classes with the accuracy all above 80%. These results indicate that the proposed method can be used as an alternative to identify and extract information related to vegetation cover for large areamapping activities.</em></p><em><strong>Keywords: </strong>OBIA, large area, land use cover change (LULC), landsat image, decision tree</em>