dc.description.abstract | Rice is the staple food in Indonesia. Information on rice crop growing areas is useful for relevant agencies to devise better strategies to ensure security and stability of national food. The main objectives of this study are to map rice cropping systems and estimate the rice yield using normalized difference vegetation index (NDVI) generated from moderate resolution imaging spectroradiometer (MODIS) data in the northern part of West Java Province, Indonesia. This study area is one of the key rice producing regions in Indonesia. It consists of nine regencies, covering approximately 13,155 km2.
The data were processed for 2008 and 2011. The methodology of this study comprises two main parts: rice crop mapping and rice yield model establishment. To map rice area, four steps need to be performed: (1) data preprocessing to construct smoothed time-series MODIS vegetation indices, including: NDVI, land surface water index (LSWI), normalized difference built-up index (NDBI); using wavelet transform, (2) image masking to eliminate non-crop areas, (3) rice crop classification using the supervised Support Vector Machine (SVM) classifier, and (4) classification accuracy assessment using the ground reference data and rice area statistics. To establish a rice yield model, three steps were needed to be performed: (1) heading date NDVI extraction, (2) rice yield modeling, and (3) error analysis of rice yield model.
The classification results showed that the overall accuracy and Kappa coefficient of MODIS-derived rice area in 2008 are 86.5% and 0.73 respectively, while in 2011 are 86.4% and 0.728 respectively. For rice yield estimation model established from this study, several models (one, three, five, and seven points of NDVI value in heading period) were generated and the three points rice yield model was chosen as the best model since it represent the shifted period of planting date for the farmers in the study area. The rice yield model showed high correlation coefficient (R2) of 0.72 and RMSE of 0.418 ton/ha. The study results indicated that the rice crop mapping and yield estimation model using MODIS imagery was reliable and reasonable. | en_US |