Advancing agroecosystem modelling of nitrogen losses with machine learning

Minimising nitrogen (N) losses from agricultural practices is critical to address the challenges of environmental degradation and climate change, and safeguard global food production systems. Agroecosystem simulation models are often used to estimate how much, and through which pathways, N is lost to the environment so that better fertiliser management decisions can be made. However, traditional process-based models, while foundational, often struggle with the calibration and validation processes across the expansive datasets originating from diverse agricultural settings. With agriculture moving into the big data era, machine learning (ML)-based data analytics presents an exciting opportunity to make a breakthrough in advancing process-based agroecosystem models. Here we explore the limitations of process-based models in predicting agricultural N losses and discuss the potential of ML-based approaches to overcome those challenges. We propose a hybrid model framework that combines process-based and data-driven ML models, offering a novel approach that leverages the strengths of both to enhance the accuracy and scalability of N loss prediction in agriculture.