Tehnik mendeteksi lahan longsor menggunakan citra spot multiwaktu: studi kasus di Teradomari, Tochio dan Shidata Mura, Niigata, Jepang

Abstract

This study describes the use of multitemporal principal component analisys (MPCA), vegetation index differencing (VIDN) and conventional maximum likelihood classifier (MLC) for detecting landslides. The study found that the synthetic images derived from stable greenness, delta greenness and delta brightness of MPCA summarized the information of landslides effectively producing accuracy of 88% for Teradomari and 91% for Tochio and Shitada Mura. The VIDN provides relatively lower accuracies than those from MPCA, i.e., only 62.5% for Teradomari and 64% for Tochio and Shitada Mura. The MLC method also provided very low user’s accuracy, i.e. 56.9% for Teradomari and 63.7% for Tochio and Shitada Mura but high producer’s accuracies, i.e.100% for Teradomari and 98.3% for Tochio and Shitada Mura. The study also found that the landsides that could be detected should be more than the size of spatial resolution of the SPOT imagery, i.e. 10 m x 10 m. Detecting landslides using SPOT imagery is more efficient than using only ground survey, providing an efficiency of 2.7.