OPTIMUM TIME LAG DETERMINATION FOR YIELD MONITORING WITH REMOTELY SENSED IMAGERY

Abstract

The time it takes for the grain to pass through the mechanisms of a harvester before reaching the point ofmeasurement by a yield monitor greatly affects the accuracy of maps generated from yield monitor data. This time lag isgenerally adjusted based on limited field observations. This article presents a method for determining time lag for yieldmonitoring using remotely sensed imagery. This method is based on the assumption that an incorrect time lag will cause areduction in the correlation between yield and remotely sensed imagery of the field. Therefore, a time lag that maximizes thecorrelation can be considered the correct or optimum time lag. To illustrate how this method works, yield monitor data andairborne multispectral imagery collected from a number of grain sorghum fields in 19982000 were used. The yield data wereassociated with their geographic coordinates with time lags ranging from 0 to 25 s at onesecond intervals. The airborneimages were georeferenced to the same coordinate system as the yield data and resampled to 1 m pixel resolution. Both theyield data and the images were then aggregated into grids with a cell size equivalent to the combines effective cutting width.Regression analyses were used to calculate coefficients of determination for equations relating yield to each of three imagebands and four vegetation indices (two band ratios and two normalized differences) and to a combination of the three bandsfor each of the 26 time lags. The plots of coefficients of determination against time lag for each of the seven spectral variablesand the combination of the three bands were bellshaped. The time lags corresponding to the maximum coefficients ofdetermination for all the curves of the spectral variables were essentially the same for a given field in a given year. Moreover,optimum time lags determined from images collected from a given field at different dates during a growing season were almostthe same. For yield monitor data collected with sampling intervals larger than 1 s, a bellshaped model and a quadratic modelwere fitted to the curves to accurately estimate optimum time lags. These results show that optimum time lags can bedetermined from either individual image bands, all the bands combined, or vegetation indices derived from airborne imagerytaken during the growing season.