Spotlight on the impact of climate change on agriculture

Building the UK’s Crop Modelling Framework

Agriculture has important socio-economic implications for creating food security by ensuring the availability, accessibility and affordability of nutritious food for everyone. Since the production patterns of crops rely on specific climate conditions, agriculture is severely vulnerable to climate change. Agriculture is also a direct cause of 8.5% of greenhouse gas emissions (IPCC report), while current agriculture practices additionally constitute about 15% of emissions from land use change, such as deforestation or land clearing for food production. Further exploitation of agricultural land through industrial mono-planting, over-irrigation and use of harmful chemicals to control evolving crop diseases has resulted in deteriorating soil health. It is hardly a surprise that the United Kingdom Food Security Report (2021) identifies climate change, alongside the loss of biodiversity and misuse of natural capital resources as the biggest threats to food security. In coming years, likely, the current food system in the UK will no longer be viable to meet consumer demand.

Under the ‘impact of climate change on agriculture’ project, the Turing researchers are building a crop modelling framework for the UK in collaboration with Rothamsted Research, John Innes Centre, University of Exeter/Met Office, The National Phenomics Centre at Aberystwyth University and the Centre for Ecology and Hydrology (UKCEH). Turing’s researchers are finding optimum ways to combine data and models efficiently, and using these insights to inform the development of the Scivision tool along with knowledge gained from the DeepSensor and Cryo-em projects, as well as building tools to extract more detailed data on plant development. Multi-disciplinary information from plant science, hydrology, soil science, insect population dynamics, economics, consumer behaviour, plant pathology, crop yields and climate models are being integrated into models of plant development and crop yield and scaled up to make nationwide predictions to support the new policy and management practices.

AI applications are embedded to widen our understanding of agricultural aspects including plant breeding conditions, management practices such as fertiliser treatment, as well as crop protection measures against diseases, weeds and pests. To build knowledge about disease and climate-resistant crops, researchers are integrating crop yield data and disease information from Agriculture and Horticulture Development Board (AHDB), and weather variables from the UK’s Met Office. By further incorporating satellite data and soil data researchers are constructing large-scale training data sets in collaboration with the UKCEH. These complex datasets are logically filtered and combined for exploratory visualisations and subsequently analysed to understand the association between meteorological and soil variables and the pathogen-associated yield gaps. This training data has been rigorously aligned on various temporal and geographic scales and further anonymised of commercially sensitive data from UK farmers – making it a significant output from this project.

Integration of plant protection data into crop models has already been achieved for five major AHDB crop varieties (winter wheat, spring wheat, winter barley, spring barley, and spring outs) (Raza and Bebber 2022). It predicts future yields based on the levels reported by AHDB and characterises pathogen-associated yield gaps. The gold standard crop simulation model, APSIM, can reproduce historic yield data predicting mean yield accurately. To also account for the variability of observed crop yields, An R codebase of optimization frameworks is also being developed at John Innes Centre, with data integration aided by the Turing to calibrate the most important parameters of APSIM and improve prediction of variability in crop yield. These parameters include the integration of data on temperature, water stress, and phenological development. The wheat yield models provide an exploratory analysis of long-term wheat experiment data such as from genotyping, as well as associated information on disease and weather to forecast wheat yield in the UK. In-depth modelling will be conducted to understand the wheat yield response to nitrogen application and the impact of weather conditions such as unfavourable temperatures and precipitation. The Met Office data will be explored to model fungal pathogen spore dispersal under climate change scenarios. Additional drone images of crops would improve current process-based yield models to gather longitudinal data and classify disease linkage more efficiently.

Open source AI/ML approaches are also enabling the analysis of plant images to extract phenotype data such as characterisation of the size, shape and distribution of seeds. A deep learning pipeline for the detection and segmentation of seeds in 3D X-ray CT images of seed pods was built based on an open source model StarDist and has been integrated into the Scivision tool to analyse cell microscopy images. Further application of open source computer vision methods will enable the exploration of data to better understand the architecture of oilseed plants (Corcoran et al. 2023). The project team is also finding applications for techniques developed for the plant phenotypes in analysing other image data, such as for the Living with Machine programme in collaboration with DEFRA and BBSRC. All models and data will be integrated into the Scivision and the demonstrator notebooks will be made available via the Environment Data Science Book allowing users to easily load new data and perform inference with these models.

Identifying, reconciling and integrating datasets and modelling methods, as well as exploring how machine learning tools could improve crop yield and plant development models could allow us to both predict the impact of climate change on UK agriculture and identify arable crops that are resilient to climate change. The Turing’s research in this area will contribute to ensuring future food security by helping identify ways for farmers, policymakers and local stakeholders to mitigate future risks and adapt to changing climate conditions shaping the consumption and production patterns of our society in the near future. Data-driven insights from the crop modelling framework will improve our ability to predict plant development and crop yield indefinitely, strengthening food security over many years to come.