1. PEMETAAN PENGGUNAAN LAHAN SAWAH BERDASARKAN PENDEKATAN EKOLOGI BENTANG LAHANMENGGUNAKAN CITRA PEREKAMAN TUNGGAL
- Author
-
Algi Variski Hasibuan, Projo Danoedoro, and Sigit Heru Murti
- Subjects
convolutional neural network ,landscape ecology approach ,maximum likelihood ,rice field ,single-date imagery ,Land use ,HD101-1395.5 - Abstract
A rice field land-use map is essential in the sustainable land management of rice fields for physical monitoring and planning. Such maps are usually created using multitemporal image data with a spectral approach, but this method can only be applied to certain areas and cannot be easily applied to other areas with different land characteristics. While multitemporal data has been widely used by researchers and proven effective, using single-date imagery can be more efficient. This study aimed to map rice field land-use based on a single-date Sentinel-2 imagery and landform maps. These landform maps were derived through visual interpretation of false colour composite bands, DEMNAS, and land system map. The interpretation resulted in eleven landform classes. The landscape ecology approach assumed the influence of landforms on land-use. The use of ten optical bands in multispectral classification using the maximum likelihood algorithm and convolutional neural network algorithm resulted in twelve land cover classes. The land cover map and the landform map were implemented through a two-dimensional ecological spatial relationship matrix that produced nine land-use classes. The convolutional neural network algorithm obtained an overall accuracy of 90,28% with a Kappa of 0,87. This result was better than the maximum likelihood algorithm, which obtained an overall accuracy of 86,81% with Kappa 0,83. The land-use map for the rice field class produced by the convolutional neural network algorithm had a total area of 33.686,69 ha and a mean absolute error (MAE) value of 0,0241, while the maximum likelihood algorithm produced a total area of 29.590,21 ha with a larger MAE value of 0,0343.
- Published
- 2025
- Full Text
- View/download PDF