Balancing phosphorus fertilization for sustainable maize yield and soil test phosphorus management: A long-term study using machine learning

Abstract

ContextExcessive phosphorus (P) applications to ensure enhanced crop growth and increased profitability are commonly used, leading to declined P use efficiency and high P accumulation in most cultivated soils across north-western India. However, the effects of consistently applying excessive P fertilizers on soil test P (STP) levels, P availability, and maize production outcomes are poorly understood. ObjectivesThe study aims to find the long-term impacts of different fertilizer P application rates on maize yield, agronomic P efficiency, optimal P rates for yield maximization, and predictive models for sustainable crop production and environmental protection across soils with different STP status. MethodsThe current study collected and analysed maize plant and soil data spanning from 2011 to 2021 from an ongoing long-term field experiment conducted on three different STP status soils i.e., medium P (MPS), high P (HPS), and very high P (VHPS) soils. The experiment involves five different fertilizer P application rates No-P, 13, 26, 39, and 52 kg P ha−1. ResultsThe results indicated a significant increase in maize grain yield and P bioavailability with the application of 13 kg ha−1 of fertilizer P in VHPS (>50.0 kg P ha−1) and 26 kg ha−1 in both HPS (22.5–0.0 kg P ha−1) and MPS (12.5–22.5 kg P ha−1). Improved agronomic efficiency reducing P rate to half of the recommended 26 kg P ha−1 was estimated in VHPS. The economically optimal P rate (EOPR) model consistently sustains crop yield with lower P rates. Among the different machine learning algorithms, multiple linear regression (MLR) was found more reliable in predictive accuracy. The decrease in predicted grain yield and increase in predicted soil P accumulation at P rates of 26 kg P ha−1 and above in varying P status soils underscores the need for judicious P fertilization practices to avoid negative impacts on yield. ConclusionsThe P application required to obtain an average maize yield of 6000 kg ha−1 with relatively high P use efficiency, fertilizer P applications of 35.2 kg ha−1 in VHPS, 37.1 kg ha−1 in HPS, and 41.1 kg ha−1 in MPS based on STP status is more economical. Smart agriculture, using machine learning algorithms helps in ensuring effective P utilization over the long term while minimizing negative ecological impacts. ImplicationsThe sustainable growth of maize crops will necessitate a reduction in P application rates, and P optimization should need further attention to explore the microbial dissolution potential of residual P.