Between data-science and geosciences: exploring the use of deep learning in automated mapping

Abstract

Artificial intelligence (AI) is one of the fastest-growing disciplines in information technology and has been successfully applied in many different fields. Currently, the technology is available for everyone but remains challenging due to its complexity. This Ph.D. research takes up the challenge and aims to bridge the disciplinary gap between data science and geoscience. In geoscience, traditional mapping relies heavily on manual labour, data analysis, and expert knowledge and thus could be done more efficiently using AI technology. This PhD demonstrates how non-experts can harness and integrate deep learning for mapping purposes based on the use of freely available data (e.g., terrain derivates and satellite imagery, Landsat 8, Sentinel-1, and Sentinel-2) to simplify daily tasks and to improve future work processes. Deep learning is the subset of machine learning methods based on artificial neural networks.

This article-based thesis consists of six independents. However, they are interrelated research articles that use deep learning applications for semi-automatic geospatial classification. It identifies critical geoscientific features such as outcrops, wetlands, landslides, and ravines. Deep learning proved to support and improve hazard assessment and support new mapping strategies with the use of available technology. The research highlights the importance of interdisciplinary collaboration between data scientists and geoscientists to drive innovation and maximize the potential of deep learning. Bridging the gap between expert data science knowledge and geoscience expertise, this research makes significant contributions to a more sustainable future for expert-based decisions.

Moreover, this study evolved over the last four years and identified the approach of combining predictions of multiple models as a promising approach (ensemble models).

In conclusion, the main contribution of the study shows practical implications for the geoscience community, as well as for the society at large. The application of automated mapping can lead to a better understanding of geospatial processes over larger areas and their impact on the environment.