Evaluation of Landsat 8 time series image stacks for predicitng yield and yield components of winter wheat

The accurate prediction of crop yield is of great importance to regional production and food security. Many empirical models for crop production prediction are based on vegetation indices (VI) such as normalized vegetation index (NDVI) or simple ratio (SR) in specific growing period, with little attention paid to the sensitivity of different growing stages to yield prediction. This study investigates the sensitivity of different growing seasons to wheat yield and yield components with time series stacks of Landsat 8 Operational Land Imager (OLI) images over the growing season of winter wheat in a farm of Jiangsu, China. The final yield and yield components including panicles per m
²
, grains per panicle and 1000-grains weight of winter yield were estimated using three regression methods namely simple linear regression, stepwise multiple linear regression and regression tree. Our experimental results demonstrated that the best estimation of wheat yield was produced using the regression tree method with a R
²
of 0.87. Among the three yield components, the 1000-grains weight was best estimated from Landsat data acquired at early stages. The estimations of specific yield components from Landsat data are useful for us to better understand the prediction of wheat yield with time series satellite image data.