Artificial Intelligence utilizing Space assets for Science discovery

2017-2027年空间地球科学应用十年调查发现1.1指出，基于空间的地球观测提供了地球的全球视角，改变了我们对地球的“科学理解”，空间的有利位置使我们能够看到地球不断变化的过程对我们生活的影响程度。然而，地球观测卫星每天产生的数据量太大，人类无法消化、分析和综合成有意义的决策和战略，以缓解气候变化和备灾。将人工智能（AI）与空间资产相结合，可以通过揭示新的联系，帮助自主决策，提高对生态系统之间复杂关系的科学理解，最大限度地提高空间任务的科学回报。目的和目标这项博士研究的目的是利用天基资产进行与气候减灾、备灾和应对有关的科学研究，并通过使用人工智能帮助科学理解。本研究将探索深度学习，通过卷积神经网络架构从高维数据中提取信息，从时间序列EO数据集中学习时空特征，从而揭示以前未知的关系。本文还将探讨模型的可解释性和不确定性量化。可解释性很重要，因为能够有效地将模型如何生成预测传达给利益相关者、政策制定者和政府是至关重要的，因为如果不清楚地理解，这些实体不太可能采用这些可靠的人工智能解决方案。研究方法的新颖性本研究的新颖性将是人工智能和机器学习模型的识别和制定，这些模型可以跨科学领域和用例应用。对于EO领域的人工智能研究人员来说，关注模型的可解释性、可解释性和不确定性是探索的开始阶段，这对于将这些技术应用于现实世界至关重要。与EPSRC的战略和研究领域保持一致本研究与EPSRC的多个研究领域保持一致，即人工智能技术、运筹学和英国气候适应计划。任何公司或合作者参与博士工作将由Yarin Gal教授和高级研究员Freddie Kalaitzis监督。通过工业学生与卫星应用弹射器和火卫二空间的合作将提供EO专业知识来支持研究

Brief description of the context of the research including potential impactThe 2017-2027 Decadal Survey for Earth Science Applications from Space finding 1.1 states that Space-based Earth Observations provide a global perspective of Earth that has transformed our "scientific understanding" of our planet, and the vantage point of space enables us to see the extent to which Earth's ever-changing processes influence our lives. However, the volume of data generated daily by Earth Observation (EO) satellites is far too great for humans to conceivably digest, analyze, and synthesize into meaningful decisions and strategies for climate change mitigation and disaster preparedness. The use of Artificial Intelligence (AI) in combination with space assets can maximize the scientific return of space missions through revealing new connections, aid in autonomous decision making, and improve scientific understanding of complex relationships between ecosystems.Aims and ObjectivesThe aims of this PhD research are to utilize space-based assets for scientific research related to climate disaster mitigation, preparedness and response, and aid in scientific understanding through the use of AI. Deep Learning will be explored in this work, which can unveil previously unknown relationships through extracting information from highly-dimensional data via convolutional neural network architectures to learn spatiotemporal features from timeseries EO datasets. Interpretability and uncertainty quantification of models will also be explored in this work. Interpretability is important as it is crucial to be able to effectively communicate how models generate predictions to stakeholders, policy makers, and governments, as these entities are less likely to adopt these AI solutions as reliable if not clearly understood.Novelty of the research methodologyThe novelty of this research will be the identification and formulation of AI and Machine Learning models that can be applied across scientific domains and use cases. The focus on model explainability, interpretability, and uncertainty is at the beginning stages of exploration for researchers within the AI for EO field, which will be critical to implementing these technologies for real-world use.Alignment to EPSRC's strategies and research areasThis research aligns with multiple EPSRC research areas, namely Artificial intelligencetechnologies, operational research, and the UK climate resilience program.Any companies or collaborators involved The PhD work will be supervised by Professor Yarin Gal and Senior Research Fellow Freddie Kalaitzis. Collaborations with the Satellite Applications Catapult and Deimos Space through an industrial studentship will provide EO expertise to support the research