A comparative analysis of multitemporal MODIS EVI and NDVI datafor large-scale rice yield estimation

Abstract

Rice is one of the most important food crops worldwide, and large-scale rice yield estimation is thus crit-ical for planners to formulate successful strategies to address food security and rice grain export issues.This study performed a comparative analysis of multitemporal Moderate Resolution Imaging Spectro-radiometer (MODIS) enhanced vegetation index (EVI) and normalized difference index (NDVI) data forestimating rice crop yields in the Mekong River Delta (MRD), Vietnam. We processed the data for a 10-year period (2002-2011) following three main steps: (1) create a smooth time series of EVI and NDVI data, (2) formulate crop yield models, and (3) validate the model. The comparison results between EVI/NDVI-based estimated yields and the government's yield statistics indicated a significant relationship betweenthe two datasets (p-value < 0.001). The estimated results produced from EVI-based models were slightlymore accurate than those from NDVI-based models, with the correlation coefficients (R²) ranging from0.62 to 0.71 for spring-winter and 0.4 to 0.56 for summer-autumn rice crops, respectively. The root meansquare error (RMSE) and mean absolute error (MAE) used to measure the model accuracy revealed theconsistency between EVI-based estimated yields and the government's yield statistics. The RMSE valuesfor winter-spring and summer-autumn crops were, respectively, 6.9-8.1% and 5.4-6.7%, and MAE valueswere 5.4-6.7% and 6.5-9.5%. There was, however, a significant correlation between the estimated yieldsobtained from EVI- and NDVI-based models (p-value < 0.001), indicating no significant difference in theestimated yields between these two models. This study demonstrates advantages of using multitemporalMODIS EVI data for large-scale estimation of rice crop yields using the heading date in the MRD prior tothe harvest period, and thus the methods could be transferable to other regions.