Detection of Rice Distribution Using NDVI Time-Series Similarity Measure

Abstract

Rice cultivation is essential for agricultural production in Mekong Delta. Rice production estimation is based on the measurement of paddy field area and plantation system. Detection of paddy rice area and the plantation system is significant to food security decision. Based on the MODIS NDVI time-series data and the Dynamic Time Wrapping (DTW) similarity measure we detected the single and multiple rice distribution in Mekong Delta in 2010. Firstly, a combination of replacement and filter method was applied to denoising the MODIS NDVI time-series. Secondly, extracted the standard NDVI time-series cycle of single and multiple rice on MODIS NDVI time-series image. Thirdly, compare each pixel’s NDVI time-series cycle with each standard NDVI time-series cycle based on DTW distance. Finally, choose threshold in DTW distance and determine paddy rice’s distribution. Results were examined by sampling point and high resolution image of Mekong Delta; it shows that the extraction accuracy is 86%, relatively high for MODIS data classification outcome.