Agriculture is among the most interconnected sectors on the planet.It is critical to alleviating poverty: about 78% of the world's poorest people live in rural areas and rely on agriculture for subsistence (World Bank).Agriculture is key in climate change mitigation: according to the United Nations Food and Agriculture Organization (FAO), agriculture accounts for about 25% of global emissions and occupies about 38% percent of global land surface. Agricultural policy is central to equity and justice: one in four (about 1.9 billion people) are food insecure and that number is growing due to shifts in the pandemic and conflict in major food producing regions. Agriculture is essential to health: diet related disease is the #1 cause of death globally and half of all deaths of children under 5 are attributed to hunger. As a person and scholar, I am wholly convinced that work in food systems has a very high return on investment for the wellbeing of people and the planet. Given the importance of agricultural systems, threats to agriculture from poor management, climate change and shifting global food systems are among the greatest risks to society and the environment. In the same vein, there is incredible potential for improvements in boosting yields, feeding a growing population, sparing land, conserving water, preserving soils, improving environmental health and supporting communities to thrive. Historically and currently, individual farmers and communities have monitored and managed agricultural systems at the field and local level. The regional scale of policy changes, conflict and localized weather pattern shift and the global scale of climate change renders appropriate the work of large-scale spatial monitoring and management of agriculture by region-, nation- and global- level organizations. Understanding agricultural systems at these scales relies on field work, aggregate statistics like reported yield 5, and remotely collected data. Field work and reported statistics can be time and resource intensive and not always appropriate for applications needing a large spatial or temporal range. In contrast, satellite-based remote sensing offers global coverage of land surface data beginning in the 1970s at low cost. Remote sensing data can be used for tasks like identifying crop types and estimating crop yield. The increasing availability of compute and declining cost of satellite remote sensing has made remote sensing increasingly viable for knowing about agriculture across time and space. Despite the huge potential, remote sensing of agriculture has been largely underutilized to date. Most methods for using multi-spectral timeseries remote sensing data for applications in agriculture rely on basic machine learning (ML) algorithms such as random forests (RF) and support vector machines (SVMs). These ML methods have largely existed since the 1980s and 1990s and persisted; they are easy and relatively "cheap" computationally to run and work well enough that some methods - random forests in particular - are often the baseline that other methods are compared against today. Radiative Transfer Models (RTMs) likewise began in the 1980s to provide a physically-based description of the interaction between electromagnetic radiation and plant canopies, and continue to be used today.Within the bounds of my PhD, I aim to evaluate the effectiveness of SSL for applications specifically to agriculture by the extent to which tested methods can separate the biophysical signals of soil and vegetation in spectral timeseries. Considerations of the limitations and risks of remote sensing are a fundamental part of this research and its framing as to mitigate the social and environmental risk posed by the use of remote sensing methods. I aim to bring perspectives from feminist geography into my approach to remote sensing to create more equity- and dignity- oriented framings of remote sensing of agriculture.

Global Climate Models (GCMs) are important for our understanding of Earth's past, present and future climates. They are based on fundamental physical processes, therefore accurate cloud modelling is important, with clouds strongly influencing the transfer of radiant energy and spatial distribution of latent heat in the atmosphere, influencing weather and climate. The representation of clouds is a major source of uncertainty in climate models and the primary reason behind the inter-model variance in climate response. I will be bringing probabilistic machine learning approaches, such as Gaussian Processes and Generative Adversarial Networks, to this problem. These will be used to reduce computational requirements of GCMs, reduce errors through better understanding of the distribution and uncertainty associated with cloud processes, and improve generalisability for future climate scenarios.

Space weather is a potent natural hazard capable of substantial economic damage to the satellite industry, power outages and harm to human health. Society's increasing dependence on satellite technology for environmental monitoring, navigation, communication, and defence purposes means robust forecasting of this phenomena is of high importance to society. Research will be conducted into providing forecasts using solar wind data and solar imaging using machine learning and artificial intelligence techniques. Space weather, added to the UK national risk register in 2012, is the complex interaction of the solar wind with Earth's geomagnetic system. Extreme events on the sun, such as coronalmass ejections and solar flares, can result in powerful fluxes of particle towards the Earth. This activity can cause a range of disruptions to crucial satellite services from momentary disruptions to a complete loss of a satellite, such as the $640million ADEOS II environmental research satellite in the Halloween storm of 2003 [1]. They can also cause power outages on Earth - the same storm caused a power outage in Sweden for 1 hour. Providing machine learning approaches to forecast this activity will help society deal with this potent natural hazard. Machine learning techniques will be researched to forecast indices and the state of electron radiation belts fromsolar wind data. As mentioned in previous works in the field such as Sexton et al., 2019 [3], machine learning models struggle to perform on the extreme storm event values due to their rarity. Techniques focusing on these rare events will be developed and quantified. Furthermore, probabilistic techniques will be developed to provide uncertainty measurements in these forecasts furthering works such as Chakraborty et al., 2020 [4]. It is also suggested that encoding physical laws into a model and then calibrating the model using data will yield better results than data-driven approaches alone [2]. A study into how to apply such 'grey-box' approaches to space weather forecasting will be conducted. Solar wind parameters, and by extension geomagnetic activity, should be forecastable using solar images. Little work has been published in using the algorithmic consumption of solar images for space weather forecasting purposes. Upendran et al (2020) [5] achieve some success using pre-trained convolutional networks using the 211nmwavelength at daily resolution. Work will be conducted to further this line of research, by looking at other wavelengths and using 10 years worth of data at higher resolution. Space weather is a potent natural hazard that has the potential for severe economic damage. Research will be conducted into state-of-art machine learning methods to forecast space weather conditions using solar wind and solar image data to mitigate this risk. References [1] "Geomagnetic Storms." CENTRA Technology, Inc. 2011. [2] "The Challenge ofMachine Learning in Space Weather: Nowcasting and Forecasting" Camporeale, E. American Geophysical Union. 2019. [3] "Kp forecasting with a recurrent neural network". Sexton et al. Journal of Space Weather and Space Climate. 2019. [4] Probabilistic prediction of geomagnetic storms and the Kp index. Chakraborty et al. Journal of SpaceWeather and Space Climate. 2020. [5] Solar wind prediction using deep learning. Space Weather,18,e2020SW002478 Upendran, V., Cheung, M. C. M.,Hanasoge, S., Krishnamurthi, G.(2020)

Heart disease is the most common cause of death in patients with diabetes mellitus, and the combination of type 2 diabetes mellitus (T2DM) and coronary artery disease (CAD) is a major cause of premature cardiovascular morbidity and mortality. Therapy to improve glucose metabolism by the heart have not been widely adopted as they are cumbersome, offer limited benefit and have therefore not been widely adopted in clinical practice. This proposed research builds on existing funding from the British Heart Foundation and the MRC to assess the effect of metabolic factors to improve the ability of the heart to tolerate the effects of a reduced blood supply due to coronary artery disease (ischaemia, clinically recognised as angina). A peptide, glucagon-like peptide 1 (GLP-1), that is secreted mainly by upper intestine in response to food has recently been found to improve the action of insulin in patients with T2DM and improve glucose metabolism. A number of drugs that limit the breakdown of this peptide have now been licensed to treat T2DM, and the proposed research will investigate whether increasing the level of GLP-1 in the blood improves the ability of the heart to tolerate ischaemia. Our preliminary results suggest that GLP-1 does indeed protect the heart against contractile dysfunction that occurs both during and after ischaemia. I addition, pilot studies in a small number of patients undergoing coronary angioplasty and stenting suggest that an infusion of GLP-1 protects the heart during the therapeutic procedure, which is particularly important in those with T2DM. The research will be informative by allowing the efficacy of a safer metabolic therapy to be tested and confirmation of efficacy in patients with T2DM will provide novel information about the mediation of a therapeutic effect and provide the basis for subsequent interventional studies in patients with CAD and T2DM.

The motivation behind this proposal is best summarised in the words of Thomas Alva Edison: Until man reproduces a blade of grass Nature will laugh at our so-called scientific knowledge. This proposal is part of a long-term effort aimed at reproducing Nature s remarkable ability to generate molecular machines that perform at levels near perfection. Most biological processes involve the action of enzymes, which are central for explaining the workings of living cells and organisms. However, there is no comprehensive quantitative understanding of enzyme action and our current understanding certainly fails the most severe test ? that of producing catalysts with rates that rival natural enzymes. We want to address this complex question by evolution experiments in the test tube, harnessing the forces of Darwinian evolution. The evolutionary snapshots should then tell us how a catalytic machineries involved in signalling cascades and the biosyntheis of un-natural natural products (as potential drug candidates) are built up ? and generate useful reagents that may make it possible to probe signaling pathways and make new potential drug candidates.