Estimated rice cultivation date using an extended Kalman filter on MODIS NDVI time-series data

Abstract

Rice cultivation date estimation based on remote sensing data is critical information to evaluate the damages in rice fields from natural disasters. In this study, the 8-day composite normalized difference vegetation index (NDVI) derived from the Moderate Resolution Imaging Spectroradiometer (MODIS) data was modeled as a triply modulated cosine function, and the extended Kalman filter (EKF) is used to estimate the mean, amplitude and phase parameters of the cosine function. The cultivation dates are estimated as the date where the seasonal variation derived from the EKF is greater than a threshold after its minimum. From the experimental results, the estimated cultivation dates derived from the proposed algorithm agree with rice cultivation information from the National Rice Department. The 75.56 percentages of the estimated cultivation dates is within 16 days for the rain-fed rice areas, and more than 80 percentages of the estimated data is within 16 days for irrigated areas with two crop cycles per year.