Advancing phosphorus management and fertilizer recommendations for crop production

Abstract

Phosphorus (P) fertilizer recommendations are fundamental to crop production because they influence productivity, nutrient use efficiency, farm profitability, and long-term soil fertility. However, developing reliable recommendations remains challenging because crop response to P varies substantially across environments and because uncertainty associated with agronomic response models is rarely incorporated into fertilizer recommendation frameworks. In addition, P management decisions require consideration of multiple agronomic responses beyond grain yield, including nutrient uptake, fertilizer use efficiency, economic return, changes in soil fertility, and residual fertilizer effects. The overall objective of this dissertation was to improve P fertilizer recommendations by integrating Bayesian statistical methods with multi-environment agronomic research. First, Bayesian inference was used to estimate critical soil-test P values from agronomic response models. Bayesian methods provided probabilistic estimates of critical soil-test values and more informative uncertainty estimates under limited data conditions, while Bayesian model averaging incorporated uncertainty arising from both parameter estimation and model selection. Second, these statistical approaches were extended to soil-test correlation and calibration to develop updated fertilizer recommendation functions based on multiple soil P tests while accounting for model uncertainty among candidate response models. Third, P fertilization responses were evaluated across 34 corn site-years in Kansas by integrating grain yield, P uptake, grain P removal, fertilizer use efficiency, economic return, and changes in post-harvest soil-test P. Results showed that grain yield responses occurred in only a subset of environments, P use efficiency generally declined with increasing fertilizer rate, economic optimum fertilizer rates depended on fertilizer and grain prices, and increasing fertilizer rates progressively increased residual soil-test P. Finally, soybean response to residual and direct P fertilization was evaluated across six site-years. Direct P application produced greater yield response per unit of applied P, reached maximum yield at lower fertilizer rates, and generated a greater frequency of yield response than residual P. In contrast, residual fertilization more effectively maintained soil-test P across the corn-soybean rotation. Collectively, this dissertation demonstrates that improving P fertilizer recommendations requires both robust statistical methods that quantify uncertainty and comprehensive agronomic evaluation across diverse production environments. By combining Bayesian inference, multi-model approaches, and field experimentation, this research provides a framework for developing P fertilizer recommendations that account for statistical uncertainty, environmental variability, agronomic response, and economic conditions.