H2YOLO: plant water status from cellular to whole plant level
Lead Research Organisation:
University of Nottingham
Department Name: Sch of Biosciences
Abstract
Climate change projections indicate increased warming and drying trends for the upcoming decades. As a result, water availability is expected to become increasingly limiting for our agricultural systems. During the process of photosynthesis, plants use pores on the surface of their leaves to take carbon dioxide from the surrounding environment. However, at the same time, water is lost through the process of evapotranspiration. Water use efficiency (WUE) refers to the balance between these two processes. Optimising WUE is a key target for crop improvement, with the need to reduce water inputs (i.e. irrigation) to our cropping systems.
Water impacts plant growth across all scales; from cellular to whole plant level. At the cellular scale, stomata control the exchange of gases, whilst physical structures and intercellular channels control water movement through plant tissues. The number and shape of the stomata therefore determine the gas exchange capacity. When water exits the leaf, the turgor pressure within the plant tissue changes, causing the leaf to change shape. For example, under low water availability, a leaf may droop or wilt. Therefore, water loss is able to alter whole plant structure, and, conversely, quantifying these structural changes can be used to infer water use.
At present, rapid and reliable methods to determine cross-scale plant water use are lacking. This will be overcome within this project through the application of deep learning: computational methods that enable a computer to be 'taught' to complete a task, in this case identifying and characterising structural features of plant. By partnering researchers from the UK at the University of Nottingham and the University of Essex, with researchers in Australia at the University of Adelaide, we will explore how plants respond to water availability in diverse environments.
The aims of H2YOLO are to:
1. Use handheld microscopes and cameras to capture photos and videos of plants, focusing on cellular structures, leaves and whole plants
2. Determine how stomata are distributed across a leaf surface by using deep learning for detection
3. Quantify how transpiration impacts leaf movement by detecting and tracking leaf movement
4. Compare water use across contrasting plant species, including the model plant Arabidopsis thaliana and commercial varieties of grapevine (Vitis vinifera),
To achieve this requires developments in a number of key areas. This includes strategies for capturing photos and videos of plants, as well as improvements to the deep learning process.
Through this project, we will showcase and advance the application of AI (i.e. deep learning) methods for bioscience disciplines. We will produce new datasets and analysis pipelines that support biological and computational research, which will be shared publicly. Ultimately, this research will support the identification of plants with improved WUE, and thus enable improvements to crop yields under climate change.
Water impacts plant growth across all scales; from cellular to whole plant level. At the cellular scale, stomata control the exchange of gases, whilst physical structures and intercellular channels control water movement through plant tissues. The number and shape of the stomata therefore determine the gas exchange capacity. When water exits the leaf, the turgor pressure within the plant tissue changes, causing the leaf to change shape. For example, under low water availability, a leaf may droop or wilt. Therefore, water loss is able to alter whole plant structure, and, conversely, quantifying these structural changes can be used to infer water use.
At present, rapid and reliable methods to determine cross-scale plant water use are lacking. This will be overcome within this project through the application of deep learning: computational methods that enable a computer to be 'taught' to complete a task, in this case identifying and characterising structural features of plant. By partnering researchers from the UK at the University of Nottingham and the University of Essex, with researchers in Australia at the University of Adelaide, we will explore how plants respond to water availability in diverse environments.
The aims of H2YOLO are to:
1. Use handheld microscopes and cameras to capture photos and videos of plants, focusing on cellular structures, leaves and whole plants
2. Determine how stomata are distributed across a leaf surface by using deep learning for detection
3. Quantify how transpiration impacts leaf movement by detecting and tracking leaf movement
4. Compare water use across contrasting plant species, including the model plant Arabidopsis thaliana and commercial varieties of grapevine (Vitis vinifera),
To achieve this requires developments in a number of key areas. This includes strategies for capturing photos and videos of plants, as well as improvements to the deep learning process.
Through this project, we will showcase and advance the application of AI (i.e. deep learning) methods for bioscience disciplines. We will produce new datasets and analysis pipelines that support biological and computational research, which will be shared publicly. Ultimately, this research will support the identification of plants with improved WUE, and thus enable improvements to crop yields under climate change.
Publications
Burgess A
(2024)
Double trouble: Compound effects of heat and drought stress on carbon assimilation
in Plant Physiology
Burgess AJ
(2024)
SOS: speed of stomata opening and closing is influenced by vapor pressure deficit.
in Plant physiology
Gibbs JA
(2024)
Application of deep learning for the analysis of stomata: a review of current methods and future directions.
in Journal of experimental botany
| Description | Project H2YOLO aims to develop new tools and methodologies to analyse water use by plants, from the cellular to full plant level. Centralised in state-of-the-art for artificial intelligence, this project is developing deep learning (DL) models- models that replicate activity of the human brain- to speed up the critical bottleneck that exists in analysing the phenotype of plants. We first performed a review using a systematic approach to analyse the literature and identify all previously published DL-based methods for analysing plant stomata. Stomata are pores on the surface of plant organs surrounded by specialised cells called the guard cells which act as gatekeepers, controlling the exchange of gases between the plant and the atmosphere. Whilst they aim to maximise carbon dioxide intake for photosynthesis, they must also limit the loss of water vapour therefore analysing their morphology and distribution is critical in understanding plant water use. This review indicated 42 papers published in application of deep learning for stomatal analysis. These cover various DL approaches, including object detection and semantic segmentation, and discuss their applications in different plant species. They highlight challenges like data availability and model generalisation, while advocating for collaboration between biologists and computer scientists. The paper encourages open data sharing to enhance research and improve crop monitoring in the face of climate change. We more recently published a method called StomaGAN, a deep learning tool called a Generative Adversarial Network. Through modifications to previously published GAN models which are notoriously difficult to train, StomaGAN can create realistic fake images, improving detection accuracy of DL models and reducing the need for time-consuming manual data collection. This innovation can enhance crop monitoring and climate resilience by making plant research more efficient and increasing the size of image datasets; a critical factor limiting in developing high-throughput automated analysis pipelines. Through this work we also generated three new image datasets which have been made publicly available along with the model code. During a trip to our international partner at the University of Adelaide in September, we collected a series of leaf impressions from Almond trees. These were from an trial that has been ongoing for decades, but very little is known about the different varieties of Almond or how they differ in terms of water use. When analysed, we expect this information to be used to identify the promising almond varieties to prioritise in breeding for water use. We are currently performing experimental work using stomata mutants in both Arabidopsis and rice. These mutants have different numbers and distribution of stomata and it this has frequently been identified as a potential route to breed water use efficient crops. However, preliminary results suggest that low density mutants are not more water use efficient due to water loss at night. This could reshape the whole research field to identify new targets and so further experiments will be performed over the coming months. |
| Exploitation Route | Academically, the findings encourage further collaboration between AI and plant science, helping to refine deep learning models for the analysis of water use and other physiological factors in plants. Researchers can use StomaGAN to generate synthetic datasets, improving the accuracy and scalability of deep learning approaches for plant phenotyping. The publicly available datasets and model code provide a foundation for future studies, supporting efforts to optimise crop water use. The ongoing work with Arabidopsis and rice stomatal mutants could reshape plant breeding strategies, guiding research toward more effective targets for improving water use efficiency. Beyond academia, this work has direct agricultural applications, particularly in crop breeding and precision farming. By integrating AI-driven analysis with long-term field trials, breeders can identify and prioritise drought-resistant plant varieties, such as promising almond cultivars. Companies developing agricultural technology can build upon these deep learning models to enhance automated crop monitoring, improving yield predictions and water conservation strategies. Policymakers and climate researchers can also leverage these insights to develop sustainable farming policies. Open data sharing ensures that both researchers and industry professionals can accelerate innovation, bridging the gap between AI, plant science, and practical agricultural solutions in the face of climate change. |
| Sectors | Agriculture Food and Drink Digital/Communication/Information Technologies (including Software) |
| URL | http://www.stomatahub.com |
| Title | Datasets of leaf surface impressions |
| Description | The bespoke website generated for the project hosts newly published public datasets of leaf surface impressions collected during the project |
| Type Of Material | Database/Collection of data |
| Year Produced | 2024 |
| Provided To Others? | Yes |
| Impact | Since publication on the website, many international researchers and postgraduate students have contacted the named supervisor for assistance with their own image analysis tasks. |
| URL | https://www.stomatahub.com/ |
| Title | Modified Generative Adversarial Network for image generation of stomata |
| Description | A newly modified generative adversarial network (GAN) has been created and released to generate artificial image data for use in plant phenotyping. |
| Type Of Material | Computer model/algorithm |
| Year Produced | 2025 |
| Provided To Others? | Yes |
| Impact | This model advances state of the art in providing solutions to common issues with GANs. It come with 3 new image datasets and provides a critical advance to reduce the bottleneck in plant phenotyping and the training outcomes and performance have all been released. |
| URL | https://www.comet.com/drjonog/stomagan/view/dy74VJ43r98PW1T51Jlug8Xfx/panels |
| Description | Collaboration with Andrew Merchant- University of Sydney |
| Organisation | University of Sydney |
| Country | Australia |
| Sector | Academic/University |
| PI Contribution | We hosted Professor Andrew Merchant at Nottingham for a 3 days visit to discuss collaborations and funding opportunities. We are currently preparing two joint research grants following on from this visit; one started prior to the visit and another new idea generated during the visit. |
| Collaborator Contribution | we are generating two research grant proposals together |
| Impact | grant proposals |
| Start Year | 2024 |
| Description | Collaboration with viticulture team- University of Adelaide |
| Organisation | University of Adelaide |
| Country | Australia |
| Sector | Academic/University |
| PI Contribution | As an international award, Adelaide are our critical international partners. As such myself (PI) and named researcher on the grant travelled to Adelaide in September to meet with the project partners, disseminate results and collect further data for analysis. We were shown the research institution and experimental set up and decided future directions for the project. |
| Collaborator Contribution | The partners hosted myself and named researcher for our visit in September. They provide data and expertise towards the project- particularly in the form of image data from their breeding field trials in grapevine and almond, as well as scientific expertise on the project. |
| Impact | Multi-disciplinary- plant breeding, plant physiology, computer science Exchange visits, data and knowledge exchange |
| Start Year | 2024 |
| Description | Collaboration- University of Essex |
| Organisation | University of Essex |
| Country | United Kingdom |
| PI Contribution | Myself (PI) and the named researcher travelled to Essex for a dissemination event, including knowledge exchange. We have also been data sharing. |
| Collaborator Contribution | Essex shared image data with us and hosted us for a knowledge exchange event with academics not directly linked to the named project. |
| Impact | Multi-disciplinary: Plant physiology, computer science Knowledge exchange, data sharing |
| Start Year | 2024 |
| Description | Edible garden at Grenhustle festival |
| Form Of Engagement Activity | Participation in an activity, workshop or similar |
| Part Of Official Scheme? | No |
| Geographic Reach | Regional |
| Primary Audience | Public/other audiences |
| Results and Impact | Joint with Honeybee Farmacy, we designed and showcased a stall containing edible plants at the Nottingham Greenhustle festival in the centre of the city. Free plants were provided and a survey to participants on growing their own vegetables which received over 120 responses. |
| Year(s) Of Engagement Activity | 2024 |
| URL | https://www.greenhustle.co.uk/ |