Crop Harvest Forecast via Agronomy-Informed Process Modelling and Predictive Monitoring

Reliable and timely forecasts on crop harvest bring significant benefits to agri-food industries by providing valuable input to complex decisions on production planning. Useful predictions on crop harvest require continual effort by seasoned field agronomists. However, they are often scarce resources in the real-world. A feasible way to facilitate crop harvest forecast is through developing predictive models that can exploit data relevant to crop growth and automatically generate consistent predictions. To this end, this paper presents our design of a systematic and data-driven approach to supporting online forecasts on crop harvest. Underpinned by process modelling and predictive monitoring techniques, our approach can utilise crop-growth-related information from multiple data sources and progressively generate crop harvest predictions within the crop growing season. The approach has a flexible design informed by agronomic knowledge applicable to crop growth in general, and may be tailored to different crops and production scenarios. A case study with a local farming company using its real-life production data demonstrates the feasibility and efficacy of our approach.