OPTIMALISASI PARAMETER SEGMENTASI UNTUK PEMETAAN LAHAN SAWAH MENGGUNAKAN CITRA SATELIT LANDSAT (STUDI KASUS PADANG PARIAMAN, SUMATERA BARAT DAN TANGGAMUS, LAMPUNG)

Abstract

Pixel-based digital-image classification results often contain of salt and pepper effects, while the visual classification has weakness because frequently provide inconsistent results. Due to the above, this study describes "Segmentation Parameter Optimization for Wetland Mapping Using Landsat Satellite Image" with object-based classification. The main objective of this study is to find out the optimal combination of segmentation parameters for paddy field mapping. The study was carried out in two sites namely in Pariaman, -West Sumatera Province and in Tangamus, -Lampung Province using segmentation of Landsat acquired in 2008 and visual interpretation of multitemporal Landsat images acquired in 2000-2009. Landsat segmentation covers two steps, firstly segmentation to optimize the parameter of color and compactness values, secondly to optimize the segmentation scale parameter. For validation, the study used both the visual-based and quantitative-based classification results of 2005 and 2007 derived from Quickbird image. Qualitative test includes object separation and segmentation accuracy of the first step of segmentation, while quantitative test is performed using confusion matrix on the second step of segmentation. This study results show that within the combinations of parameter values analyzed, the combination of parameter color value of 0.9, shape of 0.1, compactness of 0.5, and smoothness of 0.5 provides the most similar segmentation to the data reference. Meanwhile, the best case that the rules of cartography is scale of 8 for Pariaman study area and scale of 6 for Tangamus study area having accuracy ranges from 90.7% to 96.3%. This study concluded that the effect of the uncertainty of geometry of Landsat images against Quickbird images shows the maximum error of segment tolerance the origin of 4 ha to 16.70 ha for Pariaman site and 13.32 ha for Tangamus site. This results are still acceptable in segmentation results. Finally it was found that, the most optimal combination of parameters for mapping paddy field is at a scale of 1:1, color of 0.9.