Spatiotemporal Data Models and Algorithms for Earth and Environmental Sciences
Scientific Area: Climate Science & Climate Change
MIT PIs:
Justin Solomon – Associate Professor Department of Electrical Engineering & Computer Science
PT PIs:
José Manuel Matos Moreira – Universidade de Aveiro
Abstract: Scientists and engineers working in fields such as the environmental sciences, the oceans, climate, or earth sciences have access to massive amounts of geo-referenced data. These data allow monitoring and studying the behavior of objects or events of interest over time, making diagnoses and predictions, etc. These tasks assume the existence of good quality data and methods and tools to analyze the data with little effort. Currently, there are many tools help on managing, processing, and analyzing spatial data, but the same does not happen when one intends to work with spatial data that evolves over time.
This project focuses on the development of models and tools for the processing of spatiotemporal (SPT) data, based on two case studies: environmental engineering and marine ecology. The focus will be on SPT data modeled as 2D and 3D geometries that can change position, shape, or size continuously over time (moving objects). For example, we can model an iceberg as a 3D moving object (thus representing its movement and changes in size and shape over time). This model has advantages over discrete models, particularly when one intends to represent the evolution of geometrically definable objects or events, as it allows for more compact and intuitive representations, and guarantees the independence of the data from the acquisition process. Two main topics will be investigated: 1. Research work in this area has focused almost exclusively on the modeling of 2D moving objects. In this project, we intend to take the first steps towards the modeling of 3D moving objects in database or data stream systems. In particular, we will investigate the feasibility of using a unified model to represent 2D and 3D moving objects.
The starting point will be the use of models based on 3D meshes or 3D voxel and interpolation methods well known in the field of Computer Graphics. In addition to defining the model, we will also study and develop algorithms to implement a basic set of ST operations. This research guideline is based on results from previous work where we used 2D meshes to represent 2D moving objects. 2 New SPT visualization techniques are required to deal with ST datasets increasing in size. Common visualizations may lack efficiency or effectiveness in transmitting the story of an SPT entity: dynamic visualizations are not efficient as the time effort to see the visual content increases linearly with the data. Moreover, object changes may be imperceptible for small or fast-occurring transformations. Static visualizations also lack the easiness of conveying dynamic phenomena effectively. We already proposed an approach to visually represent a SPT entity based on the automated generation of interactive storyboards that summarize the evolution of the entity. Here the most critical component is the detection and representation of change, currently supported in pairwise PSR techniques (CPD). However, a new family of PSR techniques, based on machine-learning (ML), have recently emerged, which might be an alternative to more classic techniques such as CPD, ICP and BCPD. Here we aim to study ML PSR solutions, test and compare them with already achieved results. We will develop generic solutions that can be used in different types of applications. They will be tested using real data and two case studies with considerably distinct features. The first case study consists of modeling the spread of controlled fires using data extracted from aerial images (videos) captured in 3 field experiments carried out recently. The second case study consists of modeling the evolution of 3D coral reefs based on data collected periodically. The results will be validated by domain specialists. A comparison will also be made between the data created using the model proposed in this project and the field observations, with the data generated by a well-known fire propagation simulation model (Farsite). The duration of the project is 12 months, and the strategy will consist of testing solutions and defining guidelines for future research. We will have in mind the study of solutions for the areas of databases and GIS, as well as more recent trends, namely machine learning, data stream analysis, and digital twins. The participating institutions are the University of Aveiro, INESC TEC Porto, and the Polytechnic Institute of Leiria.
The team consists of six researchers: four from the area of computer engineering and computer science, one from environmental engineering, and another from the area of marine ecology. Professor Justin Solomon, leader of the “Geometric Data Processing” group of the “Computer Science and Artificial Intelligence” laboratory at MIT, will also collaborate with this project.