Multi-sensor and data fusion approach for determining yield limiting factors and for in-situ measurement of yellow rust and fusarium head blight in cereals

Abstract

The world’s population is increasing and along with it, the demand for food. A novel parametric model (Volterra Non-linear Regressive with eXogenous inputs (VNRX)) is introduced for quantifying influences of individual and multiple soil properties on crop yield and normalised difference vegetation Index. The performance was compared to a random forest method over two consecutive years, with the best results of 55.6% and 52%, respectively. The VNRX was then implemented using high sampling resolution soil data collected with an on-line visible and near infrared (vis-NIR) spectroscopy sensor predicting yield variation of 23.21%. A hyperspectral imager coupled with partial least squares regression was successfully applied in the detection of fusarium head blight and yellow rust infection in winter wheat and barley canopies, under laboratory and on-line measurement conditions. Maps of the two diseases were developed for four fields. Spectral indices of the standard deviation between 500 to 650 nm, and the squared difference between 650 and 700 nm, were found to be useful in differentiating between the two diseases, in the two crops, under variable water stress. The optimisation of the hyperspectral imager for field measurement was based on signal-to-noise ratio, and considered; camera angle and distance, integration time, and light source angle and distance from the crop canopy. The study summarises in the proposal of a new method of disease management through suggested selective harvest and fungicide applications, for winter wheat and barley which theoretically reduced fungicide rate by an average of 24% and offers a combined saving of the two methods of £83 per hectare.