Application of Remote Sensing, GIS and Machine Learning with Geographically Weighted Regression in Assessing the Impact of Hard Coal Mining on the Natural Environment

• Main Authors: Anna Kopeć, Paweł Trybała, Dariusz Głąbicki, Anna Buczyńska, Karolina Owczarz, Natalia Bugajska, Patrycja Kozińska, Monika Chojwa, Agata Gattner
• Format: Article
• Language: English
• Published: MDPI AG 2020-11-01
• Series: Sustainability
• Subjects: machine learning; wetlands; mining subsidence; spectral indexes; SBAS
• Online Access: https://www.mdpi.com/2071-1050/12/22/9338

Mining operations cause negative changes in the environment. Therefore, such areas require constant monitoring, which can benefit from remote sensing data. In this article, research was carried out on the environmental impact of underground hard coal mining in the Bogdanka mine, located in the southeastern Poland. For this purpose, spectral indexes, satellite radar interferometry, Geographic Information System (GIS) tools and machine learning algorithms were utilized. Based on optical, radar, geological, hydrological and meteorological data, a spatial model was developed to determine the statistical significance of the selected factors’ individual impact on the occurrence of wetlands. Obtained results show that Normalized Difference Vegetation Index (NDVI) change, terrain height, groundwater level and terrain displacement had a considerable influence on the occurrence of wetlands in the research area. Moreover, the machine learning model developed using the Random Forest algorithm allowed for an efficient determination of potential flooding zones based on a set of spatial variables, correctly detecting 76% area of wetlands. Finally, the GWR (Geographically Weighted Regression (GWR) modelling enabled identification of local anomalies of selected factors’ influence on the occurrence of wetlands, which in turn helped to understand the causes of wetland formation.