A comparative study of yield monitor data processing methods for on-farm agronomic trials

Processing yield monitor data remains a challenging task in on-farm research, particularly when evaluating yield response to inputs, such as seed and nitrogen (N). This study evaluated the comparative performance of two yield monitor processing algorithms. The first is simple and is based upon empirical thresholds, and the second is RITAS (Rectangle creation, Intersection assignment, Tessellation, Apportioning, and Smoothing) and is a constructive and computationally expensive algorithm. The effects of the different processing algorithms on the model estimates were compared in a simulation study, and two case studies with experimental data were used to demonstrate the applicability of these algorithms in an on-farm setting. In the simulation study, the simple algorithm produced less precise and accurate estimates when compared to RITAS. For instance, the standard deviations of the agronomic optimum nitrogen rate and economic optimum nitrogen rate (EONR) were 83% and 51% greater when using simple, compared to RITAS. Furthermore, when estimating the EONR, the simple algorithm presented a bias of -8 $\mathrm{kg}N{\mathrm{ha}}^{-1}$. The experimental data results from a strip trial and a checkerboard trial showed differences of 24 $\mathrm{kg}N{\mathrm{ha}}^{-1}$ and 26 $\times {10}^3\mathrm{seeds}{\mathrm{ha}}^{-1}$ in the estimate of optimum agronomic nitrogen and seeding rates, respectively, between the two processing algorithms. This study indicates that the choice of algorithm for processing yield monitor data can greatly influence the estimates retrieved from on-farm experimental data. The RITAS algorithm produced more accurate and precise estimates than simple, indicating its potential for broad use in on-farm research.