Spatial deconvolution of yield meter data.

Data for yield maps can be obtained from modern combine harvesters equipped with a differential Global Positioning System and an Yield Monitoring System. Due to a delay/smoothing effect in the combine harvester the recorded yield for a location actually represents a shifted weighted average of harvested yield over a finite region along the swath. This is modelled as a spatial convolution of the true unobserved yield and an impulse response function (IRF). The assumption of Gaussian data together with a simultaneously parametric modelling of both the impulse response function and the spatial covariance function (SCF) of the unobserved yield, enables us to compute maximum likelihood parameter estimates of model parameters. The suggested procedure of simultaneously estimating the parameters of the IRF and SCF gives more precise results compared to methods of estimating the impulse response function where no spatial information is used. The inferred statistical model for the yield meter data may be used to predict the ``true'' yield on a dense grid of locations given the observed smoothed yield data resulting in more accurate yield maps than maps based on raw data.
Last modified: Thu Nov 2 11:47:52 MET 2000