PENGUJIAN MODEL PENDEKATAN PROBABILITAS BERBASIS PERUBAHAN PENUTUP LAHAN CITA LANDSAT TUNGGAL MULTIWAKTU UNTUK PEMETAAN LAHAN SAWAH

Abstract

Tests on a model of probability approach to paddy field mapping based on land cover changes that has been done in several districts in West Java showed overall accuracy an average only 65.5%. It is thought to be related to the use of annual multitemporal Landsat mosaic results so often seem illogical because the image is derived from some of the data obtained at different seasons. In this regard has been done phase II trial of the model using multitemporal Landsat-8 singles data (not mosaic data). The objective of this research is to test the validity of the probability model on multitemporal land cover changes for rice field mapping. The method used in this trial is unsupervised classification methods for mapping multitemporal land cover. Merging multitemporal land cover change in the timeframe in accordance with the date of acquisition date. Analysis of probability as a rice field area, where if two of the land cover types were detected, bare land or otherwise classified as land with probability 1, if only observed one land cover change, bare land becomes water, water or bare land is classified as a probability of 0.3. Accuracy tests using confusion matrix between the field probability image and rice field reference level 1: 5,000. The evaluation results show that the rice field probability image reached 79.7% with lowest accuracy 67.7% (Babelan Sub-districts) and the highest 86.7% (Sukamandi Sub-districts). Comparison of results with previous results showed a high significant difference with an average increase of more than 600% accuracy. Based on these results it can concluded that the probability model based on multitemporal land cover changes for rice field mapping has good accuracy.