Evaluating wheat yield response to management inputs and soil physical properties in the Western Cape province of South Africa

Abstract

Wheat (Triticum aestivum L.) production in South Africa’s Western Cape province is challenged by high input costs, spatially variable soils and erratic rainfall, indicating potential opportunities for site-specific input management. Uniform input applications often fail to account for the in-field variability, leading to inefficient input use and suboptimal yield returns. Implementing variable-rate application strategies therefore requires a clearer understanding of how soil and management factors interact to influence yield response. This study aimed to determine profit-maximising flat and variable seeding and fertiliser rates under local dryland conditions, and to assess the relative importance of soil physical and morphological properties and management factors in explaining in-field yield variability. Six field-scale on-farm trials were conducted during the 2023 and 2024 seasons as part of the Data-Intensive Farm Management project. Variable-rate management zones within fields were delineated based on observed wheat yield responses to varying seeding and fertiliser input rates. Random Forest models were used to evaluate the contribution of seeding rate, fertiliser rate and soil properties to yield variation across the fields. Profit-maximising rates were determined using direct costs, which included real-time seed and fertiliser costs together with the grain price as received in the respective season. Seeding and fertiliser rates were consistently among the strongest predictors of wheat yield, confirming that yield variability is closely linked to input management decisions. However, soil properties, such as depth for potential root development, type and depth of limiting layers, soil structure, and plant-available water capacity, also exerted significant, site-specific influences. These interactions were non-linear and differed across fields, demonstrating that optimal input management depends strongly on local soil constraints. Overall, the integration of data-intensive experimentation with machine learning provided valuable insights into how management practices interact with soil physical variability to shape yield outcomes. The findings underscore the importance of coupling precision input management with improved understanding and remediation of subsurface soil constraints to enhance productivity and resource-use efficiency in the local dryland wheat systems.