Synergising Process-Based and Machine Learning Models for Accurate and Explainable Crop Yield Prediction along with Environmental Impact Assessment

点击翻译按钮获取中文摘要

The world's rapid population growth and climate change pose challenges to sustainable food production. Agricultural crop production has long relied on Process-based models (PBMs) to forecast yields and understand how plant physiological processes interact with the environment, influencing crop growth and development. However, the PBMs suffer limitations in making accurate predictions due to complex weather/plant interactions. This is especially true for extreme events (drought, heat waves), pests, diseases, and stresses not accounted for. Process-based models' predictive abilities are hindered by uncertainties in structure, inputs, and parameters, exceeding observed yield variations over time/space.Machine Learning (ML) offers quick crop yield prediction by learning from data, but it's often a black box needing explanations. Integrating PBMs and ML has shown promise in improving predictions. Challenges remain in effective integration: choosing the right ML for accurate simulation, balancing interpretability and uncertainty. Environmental impact assessment is often overlooked.Building on our existing foundations, this partnership brings together leading researchers in agri-environment sciences, crop modelling from Germany and computer science (big data/machine learning/AI) from UK, and aims to develop an innovative AI framework by synergising process-based and machine learning models for accurate and explainable crop yield prediction coupling with environmental impact assessment. The overarching aim is to build and foster a long-term partnership between UK and Germany's top research groups to address the call theme- AI in sustainable agriculture and food and provides the added value to our ongoing research in climate-smart agriculture solutions. To achieve this, we will conduct a series of research activities including feasibility study, staff exchanges/early career researchers (ECRs) visits, facility and data access, workshops, and joint publications/funding applications. The integration of AI with agricultural modeling represents an emerging paradigm that pushes the boundaries of agricultural research. It not only offers improved crop yield predictions and climate change impact mitigation but also opens up new avenues for understanding crop dynamics, resource optimization, and sustainable farming practices. The proposed approach has the potential to be applied at different scales, ranging from individual farm fields to regional and global levels. This scalability and generalization make the AI-driven synergy suitable for addressing complex agricultural challenges and adapting to diverse environmental conditions. It has the capacity to revolutionize agriculture, leading to more efficient, sustainable, and resilient food production systems.This research offers potential benefits to farmers, consumers, policymakers, and the environment. Improved predictions will enhance agricultural decision-making, increase food security, promote climate change adaptation and mitigation, and optimize resource utilization. Additionally, the research will advance scientific knowledge and benefit industry and academic institutions.