Mid‐infrared spectroscopy and machine learning improve accessibility of Hawaiʻi soil health assessment

Abstract Monitoring soil health is important for sustaining agricultural productivity and ecological integrity around the world. However, current assessment approaches relying on conventional laboratory methods are resource intensive. Mid‐infrared (MIR) soil spectroscopy offers an opportunity to increase assessment throughput and reduce user costs, potentially improving accessibility for land managers and producers. This study aims to develop a high‐throughput, hybridized model for soil health assessment tailored to the diverse agricultural and ecological landscapes of Hawaiʻi, with potential applicability to other subtropical and tropical areas. Leveraging a newly developed spectral dataset ( n = 634) and machine learning methods, we predicted inherent mineralogy and intensive land use legacy with 94.5% and 91.4% accuracy, respectively, validated with threefold cross‐validation. Additionally, we predicted four key soil health indicators: total organic carbon (CCC = 0.97), CO 2 burst (CCC = 0.93), potentially mineralizable nitrogen (CCC = 0.89), and water‐stable mega‐aggregates (CCC = 0.79). These predicted soil features were then used to predict the Hawaiʻi soil health score. Our results demonstrate the potential for MIR spectroscopy to reshape soil health assessment in Hawaiʻi by offering a rapid, cost‐effective alternative to traditional methods. Finally, we discuss the importance of adopting a soil health testing framework to report results that are intuitive for diverse stakeholders, including local producers and land managers.