Corn growth stage estimation using time series vegetation index

Abstract

Crop growth stage is important information for decision making in many related agricultural sectors. In-time accurate estimation of crop growth stage is desired. Kernel-fitting of time series vegetation indices have shown potential in estimating crop growth stage while tolerant to noisy data and missing data. The challenge to apply such models is dealing with current year when incomplete data are available. This study proposed a progressive double sigmoid model that leverages the existing best model to compensate the incompleteness of data. The progressive double sigmoid modeling algorithm has three stages of estimation: pre-peak, early post-peak, and late post-peak. Simulation results and experiments showed that the progressive version of double sigmoid algorithm solved the problem of fitting model with insufficient data at early stages. Double sigmoid models have been compared with other alternative approaches in different treatment of data analysis. The results showed that double sigmoid models performed better than moving median window smoothing and Savitzky-Golay alone. Further studies may consider optimizing season partitions and thresholds.