Design of on‐farm precision experiments to estimate site‐specific crop responses

Abstract

Site-specific prescriptions require estimating response functions to controllable inputs across the field. The methodology of applying geographically weighted regression to on-farm precision experimentation studies opens new opportunities to study site-specific responses to inputs in farmers' fields by locally estimating the regression coefficients. However, the effect of the experiment's spatial layout, such as plot dimensions and randomization, and spatial structure of the yield response on the experiment performance are yet to be studied. Detailed information about these effects is needed to improve trial design to detect site-specific responses. A simulation study was conducted using 14,400 fields of 37 ha and 9-m resolution. Coefficients from a spatial variable response function were drawn from five random fields generated by unconditional Gaussian geostatistical simulations. Four levels of nitrogen were assigned to plots using 18 systematic and randomized chessboard designs with different plot sizes. Simulated yield data was obtained by combining the coefficients, treatment, and random error. The effect of spatial structure and the designs was assessed with measures of agreement between the true and estimated maps of regression coefficients. The ability to capture or approximate the true spatial pattern of the response function increased as the underlying response function's spatial structure increases. Overall differences in performance between design were observed across the spatial structure tested, mostly related to randomization and plot dimensions. In general best results were achieved by systematic designs with small or intermediate plot sizes (r = 0.54 ± 0.05, MAE = 0.005 ± 0.0005, SDR = 0.81 ± 0.06, and CP = 0.50 ± 0.04). Our methodology provides a path for testing designs under different spatial variability scenarios.

Bias in economic evaluation of variable rate application based on geographically weighted regression models with misspecified functional form

“…But in the Monte Carlo simulations, we can specify yield response functions and generate data from those functions, which enables us to quantify the true economic value of the misspecified GWR‐based VRA over URA and the bias in the estimation of that value in the context of the simulations. Monte Carlo simulations are commonly used in this way to gain insights into the statistical and economic properties of trial designs and estimations methods (Alesso et al, 2021; Bullock et al, 2020; Gebbers & De Bruin, 2010; McKinion et al, 2001). We conduct one thousand rounds of Monte Carlo simulation to answer our research questions.…”

On-farm experimentation practices and associated farmer-researcher relationships: a systematic literature review

“…The most frequent journals were Precision agriculture and Computers and Electronics in Agriculture. Articles from this cluster focused on the characterization of on-farm variability to refine new technologies or models (e.g., hyperspectral canopy sensing of paddy, Gnyp et al 2014), or proposed plans for on-farm precision experiments (Alesso et al 2021).…”