Exploring the potential of Multimodal Remote Sensing for Sustainable Development Goals

《2030年可持续发展议程》强调了地理空间信息和地球观测为可持续发展目标和指标提供信息的重要性。地球同步轨道与高分辨率的现场数据相结合，提供了一个机会，可以非常详细地了解和改进地球过程的动态行为，这是以前永远不可能的。多个数据源使我们能够掌握不同的信息，每一条信息都以不同的方式描述了地球表面正在发生的事情。然而，从数据分析的角度来看，合并所有这些数据集是一项具有挑战性的任务。在当前的大数据场景中，我们拥有由复杂的底层物理过程产生的海量数据，这些过程很难监控。从这些大规模的异质数据集中提取有意义的信息是一个悬而未决的问题。第二大挑战是找出这些数据集可以组合的级别-原始或像素级别、要素级别或决策级别。此外，目前的大多数数据融合技术都是基于数据驱动的人工智能(AI)方法，无法提供AI算法的透明度和可解释性。换句话说，它没有为我们提供一个明确的物理意义，即结果意味着什么，以及为什么该模型做出这样的预测。该博士项目旨在了解一些挑战，并开发数据分析方法，以利用可解释的数据融合技术来改善地球物理过程的特征。博士学位的主要研究目标有三个。首先，通过结合领域知识和最新的特征选择算法，理解为任何项目选择最相关的数据集进行有效分析的方法。第二，研究数据融合技术，以组合这些最相关的数据集；第三，探索量化融合产品的置信度的方法，从而处理与遥感产品相关的不确定性。“

"The 2030 Agenda for Sustainable Development stressed the importance of Geospatial Information and Earth Observations (EO) to inform the SDG Targets and Indicators. EO combined with hi-resolution in-situ data provides an opportunity to understand and improve the dynamic behaviors of the earth's processes in great detail that was never possible before. Multiple data sources enable us to grasp different pieces of information which each partially characterize what is happening on the earth's surface differently. Combining all these datasets, however, is a challenging task from a data analysis point of view. In the current big data scenario, we have massive amounts of data generated by complex underlying physical processes which are difficult to monitor. The extraction of meaningful information from these large-scale heterogeneous datasets is an open problem. The second big challenge is then to find out levels at which these datasets can be combined - raw or pixel level, feature level, or at the decision level. Moreover, most of the current data fusion techniques are based on data-driven Artificial Intelligence (AI) methods which fail to provide transparency and explainability of the AI algorithm. In other words, it does not provide us with a clear physical meaning of what the results mean and why the model makes that prediction. The Ph.D. project aims at understanding some of the challenges and developing data analysis methods to improve the characterization of the earth's physical processes using explainable data fusion techniques. There are three main research objectives of the PhD. First, understanding methods to select the most relevant datasets for any project for effective analysis by incorporating domain knowledge and state-of-the-art feature selection algorithms. Second, to investigate the data fusion techniques to combine these most relevant datasets, and third, to explore ways to quantify the degree of confidence of the fused product thereby dealing with uncertainty associated with remote sensing products. "