Detection of phenology using an improved shape model on time-series vegetation index in wheat

Abstract

Accurate information about the growth period can guide us to fertilize, irrigate and harvest. Much progress has been achieved in detecting the phenology with the unique features of the time-series vegetation index (VI). However, these features only reflect information about specific stages (e.g., tillering, heading, and maturity stages), ignoring the information from other important stages (e.g., jointing, booting, and filling stages). In this study, a new approach for the phenology detection of winter wheat at the whole phenological stages is described, whereby the integrated accumulated growing degree days (AGDD) combined with the shape model (SM) method (SM-AGDD) is used to detect important phenology stages of winter wheat using five classic time-series VIs derived from three sensors at the field scale. Two proximal sensors (ASD FieldSpec Pro spectrometer and a Greenseeker RT 100) and a digital camera mounted on an unmanned aerial vehicle (UAV-DC) are used to acquire the above time-series VI. The results show that the newly developed SM-AGDD with the normalized difference vegetation index (NDVI) from ASD is the best predictor of crop phenology with an average RMSE ranging from 1.0 day at maturity to 10.3 days at tillering, followed by CI, EVI, and VARI, respectively. Among the three different spectral sensors, ASD has the best performance for detecting the whole targeted stages, while UAV-DC was the worst. In particular, the accuracy of EVI has the highest improvement on all growth stages. Compared with the previous SM constructed with the days after sowing (DAS) produced by Sakamoto et al. (2010), the newly developed SM-AGDD improves the accuracy of detecting the critical stages for winter wheat phenology for all VIs. We also find that SM-AGDD has a higher accuracy to the SM constructed with accumulated photothermal time (APTT) by Zeng et al. (2016). While, it also greatly simplified the calculation. This study shows that the accuracy of the shape model method is affected by the form and characteristics of the constructed shape, which could provide the theoretical basis for accurate detection of critical phenology dates for crops.