Gan, LQ; Cao, X; Chen, XH; Dong, Q; Cui, XH; Chen, J (2020). Comparison of MODIS-based vegetation indices and methods for winter wheat green-up date detection in Huanghuai region of China. AGRICULTURAL AND FOREST METEOROLOGY, 288, 108019.

Satellite vegetation index (VI) time series data provide a feasible option for monitoring crop phenology at a large scale. However, there are limited researches that investigated the accuracy of different methods for crop phenology detection with various VIs over a large-scale region. In this study, we used four VIs, i.e. Normalized Difference Vegetation Index (NDVI), Enhanced Vegetation Index (EVI), Two-band Enhanced Vegetation Index (EVI2), and Normalized Difference Phenology Index (NDPI) derived from the Moderate Resolution Imaging Spectroradiometer (MODIS) data, combined with six methods, i.e. relative threshold at 10%, 20% or 50% of the VI's amplitude (RT10, RT20 and RT50), maxima of the curvature change rate of the fitted logistic curve (CCRmax), maxima of the first derivation of the VI curve (beta(max)), and cross-correlogram spectral matching-phenology (CCSM-P), to detect winter wheat green-up dates (GUDs) for the period of 2009-2013 in the Huanghuai winter wheat region of China. The performance of the combinations of these methods and VIs was evaluated using ground-observed GUDs from agrometeorological stations with correlation coefficient (r), regression coefficient (a), root mean square error (RMSE) and bias. We further investigated the spatial trend of residuals from a linear model between satellite- and ground-observed GUDs. Results show that NDPI outperforms the other VIs with the highest consistency with ground data in the whole region. RT10, CCRmax and CCSM-P show higher accuracy in the northern region, while in the southern region, RT20 shows relatively higher accuracy in the case of poor performance of all six methods. However, the residuals of these six methods based on NDPI show significantly positive correlations with latitude in the whole region, suggesting an uneven spatial distribution of accuracy with a tendency of underestimating GUDs at the low latitude region and overestimating GUDs at the high latitude region when applying the same method to detect GUDs over a large-scale region. It is suggested to develop a new method or combine several methods to reduce the spatial incoherence of residuals.