The Effect of Antecedence on Empirical Model Forecasts of Crop Yield from Observations of Canopy Properties

Identification of yield deficits early in the growing season for cereal crops (e.g., Triticum aestivum) could help to identify more precise agronomic strategies for intervention to manage production. We investigated how effective crop canopy properties, including leaf area index (LAI), leaf chlorophyll content, and canopy height, are as predictors of winter wheat yield over various lead times. Models were calibrated and validated on fertiliser trials over two years in fields in the UK. Correlations of LAI and plant height with yield were stronger than for yield and chlorophyll content. Yield prediction models calibrated in one year and tested on another suggested that LAI and height provided the most robust outcomes. Linear models had equal or smaller validation errors than machine learning. The information content of data for yield prediction degraded strongly with time before harvest, and in application to years not included in the calibration. Thus, impact of soil and weather variation between years on crop phenotypes was critical in changing the interactions between crop variables and yield (i.e., slopes and intercepts of regression models) and was a key contributor to predictive error. These results show that canopy property data provide valuable information on crop status for yield assessment, but with important limitations.