MODELLING BARLEY BIOMASS FROM PHENOCAM TIME SERIES WITH MULTI-OUTPUT GAUSSIAN PROCESSES

Abstract

Biomass is monitored in many agricultural studies because it is closely related to the growth of the crop. The technique of digital repeat photography that continuously capture images of a given area with an RGB or near-infrared enabled cameras, Phenocams, has been used for more than a decade mainly to estimate phenology. Studies have found a relationship between Phenocam data and above-ground dry biomass. In this context we investigate the modeling of barley fresh above and underground biomass with Green chromatic coordinate (Gcc) colour index, extracted from Phenocam data, and multi-output Gaussian processes (MOGP). We take advantage of the available very high temporal resolution data from the phenocam to predict the biomass. The MOGP models take into account the relationships among output variables learning a cross-domain kernel function able to transfer information between time series. Our results suggest that MOGP model is able to successfully predict the variables simultaneously in regions where no training samples are available by intrinsically exploiting the relationships between the considered output variables.