Estimation of Cotton Yield Using the Reconstructed Time-Series Vegetation Index of Landsat Data

Abstract

Multi-day vegetation index (VI) composite images, the basis for crop yield estimation, are subject to temporal information losses. Using the Normalized Difference Vegetation Index (NDVI) and Enhanced Vegetation Index (EVI) from Landsat_5_TM and Landsat_7_ETM, we reconstructed time-series VI using temporal information and a mathematical model to estimate yield. The results indicated the following: (i) The reconstructed model describes real changes in VI during crop growth. (ii) The Extreme model is best for cotton growth process delineation. Compared to NDVI, EVI is appropriate for the reconstruction of time series VI curves, obtaining a model-fitting coefficient of 0.97 and a root mean square error (RMSE) of 0.05. (iii) The correlation between reconstructed EVI (REVI) and yield was more significant than Landsat VI and yield, with a correlation coefficient of 0.90 during the growth stage. The determination coefficient of the model using EVI on the DOY 198 as an input variable was 0.81, and the RMSE was 481.19. (iv) The Extreme model reconstruction reduces background differences between different plots, providing the basis for regional yield estimation using remote sensing. Our method resolves issues associated with multi-day composite VI. With increases in <30 m resolution observation programs, this model is expected to be increasingly used, improving the applicability of VI-based yield estimation models.