Statistical models of yield in on‐farm precision experimentation

Abstract

On-farm precision experimentation (OFPE) is increasingly conducted using variablerate technology and precision agriculture (PA) equipment to measure the effect of changes in input application rates on yields and profits at specific fields. Classical linear regression models and new Bayesian and machine learning regressions for spatial data can be used to investigate site-specific crop response from georeferenced data. The objective of this work was to compare statistical models that can be used by researchers analyzing OFPE data to estimate crop response and better describe its spatial within-field variability. Three statistical models estimating the responses to N rates, seed rates, and site-specific soil properties from eight OFPEs were compared:(a) linear regression (LR) for spatially correlated errors, (b) Bayesian regression (BR) with random site effects, and (c) random forest regression (RF) with kriged residuals. Models were adjusted to account for spatial variation in yield response, and with and without field characteristic covariates. Modeling spatial correlation and including plot covariates improved yield predictions. Differences among methods proved to be indistinguishable with respect to average explained variance, correlation between predicted and observed values, and mean square prediction errors (PEs).However, BR and RF outperformed LR in site-specific prediction accuracy, with BR predictions having lower predictive uncertainty than RF predictions. The hierarchical Bayesian model for spatial data is a useful tool to process OFPE data, allowing direct derivation of linear coefficients and prediction uncertainty measures related to site-specific yield responses.

Assessing performance of empirical models for forecasting crop responses to variable fertilizer rates using on-farm precision experimentation

“…Random forest (RF) regression is an increasingly popular non-parametric machine learning method that has been used to model spatial data (Georganos et al, 2021;Jing et al, 2016), specifically crop response data (Mariano & Mónica, 2021;Marques Ramos et al, 2020;Paccioretti et al, 2021), where crop responses are estimated by the ensemble of regression trees using binary splits of observations based on covariate data. The ranger package in R was used for fitting and generating predictions (Wright & Ziegler, 2017).…”

On-Farm Precision Experiments (OFPE) framework: tapping local data to optimize crop sub-field scale decisions

“…Crucial to the OFPE framework was the development of ecological models that were used to characterize the relationships between the observed crop responses, experimentally varied agronomic inputs, and the remotely sensed environmental variables. Modeling the response of crops to agronomic inputs has long been a subject of agronomic research, which has not reached a consensus on one approach that adequately captures the variability across space and time for a given crop response [30][31]. Minimization of uncertainty related to the characterization of crop responses to agronomic inputs requires model selection performed on a trial by trial basis, as the form of a model appropriate for one field is not always consistent with neighboring fields, or even the same field in different years [32].…”

Assessing green manure impact on wheat productivity through Bayesian analysis of yield monitor data

“…(2011) . Intensive developments in the field of precision agriculture open the possibilities for conducting of the on-farm experiments to compare the different agronomic practices or to test the conclusions from a small-scale field experiments with advanced possibilities in data collection with a spatial resolution such as the yield monitor data throughout the standard machineries Paccioretti et al. (2021) ; Hegedus et al.…”

“…There are large number of methodological approaches for the analysis of on-farm experiments ranging from geospatial regression models to Bayesian statistical methods Kyveryga (2019) ; Cho et al. (2021) ; Paccioretti et al. (2021) ; Hegedus et al.…”

“…In contrast to small-scale field trials, the large-scale field or onfarm trials are more variable and less precise but more representative when compared with a standard agronomic practice in a given production region Piepho et al (2011). Intensive developments in the field of precision agriculture open the possibilities for conducting of the on-farm experiments to compare the different agronomic practices or to test the conclusions from a small-scale field experiments with advanced possibilities in data collection with a spatial resolution such as the yield monitor data throughout the standard machineries Paccioretti et al (2021); Hegedus et al (2023).…”