Artificial Intelligence (AI) has become a buzzword that symbolizes the next stage of innovative technological transformations and how the industry in the future will be driven. Using intelligent algorithms, data classification and smart predictive analysis, AI is useful across a large number of sectors. Geospatial AI, or Geo.AI., is a more specific subset of AI that combines the geographic data from geographic information systems (GIS) with the razor-sharp analysis and solution-based approach of AI. Geographic information systems is a conceptualized framework that provides the ability to capture and analyze spatial and geographic data. Geo.AI. can also be classified as a geographically based form of machine learning, or the study of computer algorithms. As a subfield of spatial data science, Geo.AI. uses advancements in techniques and data cultures to support the creation of more intelligent geographic information that can be useful in ocean conservation.
Ideally, the application of techniques from artificial intelligence and data science to spatial data in the earth and social sciences can be applied to ocean conservation. Recent research has shown models that specify the location of each object of interest (spatially explicit models) substantially outperform more general models when applied to spatial data, so the authors of this study asked the following questions:
(1) What exactly are spatially explicit models and what do they have in common in ecology?
(2) How can we integrate spatial and temporal aspects to various machine learning-based techniques?
(3) How much spatial data are required for these models to make a difference for ocean conservation?
Application to social sensing
Machine learning and artificial intelligence methods also have an important role to play in what is often referred to as social sensing. Social sensing can be defined as the use of user-generated digital content to better understand any given dynamics such as ocean dynamics. Social sensing has been applied to a range of tasks from identifying human mobility patterns and exploring structure in social networks, to ocean conservation and planning solutions with varying degrees of success. The process of social sensing involves the creation of data signatures (i. e. spatial, temporal, and thematic features) that are extracted from the digital trace that is left behind as people’s digital lives interact with their physical activities, all of which are obtained from the social sensing models themselves.
Datasets and reproducibility
Advancing GeoAI research requires high-quality geospatial datasets. Many AI models need to be trained on a large set of well-labeled training data. It has long been recognized in the machine learning community that the quality of models follows the ‘garbage in, garbage out’ principle, that is, a trained model is only as good as the quality of the training data. From this perspective, data are no longer merely resources to be mined by computational tools but are becoming part of the tools. High-quality datasets, such as ImageNet, have become critical for the development of new AI methods. The field of geography is fortunate to have many high-quality datasets in the public domain, such as the National Land Cover Dataset (NLCD) from the US Geological Survey and the American Community Survey (ACS) data from the US Census, not to mention the many available remote sensing images, global digital elevation models (DEM), and National Hydrography Datasets (NHD). With the change of data culture, an increasing number of companies are also sharing their geospatial data which can become useful resources for developing future GeoAI models.
Future directions speculated on by the authors
Two future directions could be explored to promote dataset and code sharing with the goal of supporting reproducibility and replicability in GeoAI research. First, we may continue to enhance our spatial data infrastructures (SDI) which serve as central platforms for sharing geospatial resources. Research efforts could be put on facilitating the search and discovery of resources on SDI, providing guidance on the best practices of data sharing, and designing automatic methods for improving the quality of geospatial data and metadata. Second, we could encourage the coupling of research articles and datasets in top journals. While this can be done by sharing a publicly accessible link of the repository within an article, an existing journal might offer a dataset track, or a new journal could be established specifically for publishing descriptions of geospatial datasets. There are already such dataset journals beyond the Geography field, such as Scientific Data published by the Nature Publishing Group. These journal articles can give more credits to researchers who spent time and effort to carefully collect, clean, and share datasets. On the other hand, new challenges need to be addressed on how to effectively review these dataset-description papers and how to ensure the quality and maintenance of the shared datasets.
Image: University of Buffalo’s Geospatial Artificial Intelligence Lab
In this editorial, the authors described the need for GeoAI research and reviewed its origins. The authors have outlined three significant research directions, namely spatially explicit models, question answering, and social sensing. Additionally, they discussed the need for high-quality datasets and improved reproducibility and presented a GeoAI moonshot as an example of a shared vision for the next ten years. They hope that GeoAI and spatial data science more broadly will combine multitude domains that work on or with spatiotemporal information. Finally, the authors believe that ethical consideration should be an essential part of responsible GeoAI research, both on the level of individual researchers as well as on the community. They believe that the breadth of topics and techniques in this paper is well representative of the current state-of-the-art in GeoAI.
Krti is interested in the transmission dynamics of environmental diseases as they relate to climate and anthropogenic stressors. As a Fulbright Scholar, Krti conducted analyses on the responses of dengue fever to climatic stressors off the coast of the Bay of Bengal, in India. Currently, Krti works with Stanford University to understand the role of schistosomiasis in environmental reservoirs, and leads the pursuit of a computational-based based analysis of eelgrass wasting disease dynamics. At Stanford, Krti serves as one of the few trans-disciplinary experts for planetary health topics, via machine learning and computer vision, data science, environmental policy, and science communication. As a STEM innovator and a first-generation woman of color, Krti is proud to be a writer for Oceanbites!