Predicting yield of barley across a landscape: a state-space modeling approach

Abstract

Spatial crop yield prediction is an enigma that needs to be solved to avoid ecological and economical risks in agricultural crop production, that can result from local fertilizer surplus or deficiency. Current approaches for site-specific fertilizer distribution are based on patterns of soil properties and yield maps obtained from previous years. The aim of this study was to evaluate the quality of crop yield prediction in an arable field using two sets of variables in autoregressive (AR) state-space models. One set included detailed soil information (texture, organic carbon content) and yield data from the previous year at a high spatial resolution. In the other set, remotely sensed soil and crop information (vegetation index, crop nitrogen status, land surface elevation) was assembled, which is available under farm conditions without intensive soil sampling campaigns. Soil and remotely sensed variables were evaluated in bi- and multivariate autoregressive state-space analysis to predict spring barley grain yield. Remotely sensed variables showed to be better predictors for spatial grain yield estimation than soil variables. Transition coefficients determined from state-space analysis were applied in AR-equations with soil and remotely sensed information, but yet given only the initial value of the spatial yield series. Both sets of variables elicited similar prediction quality.