Predicting Mining activities dynamics in Ghana: A Fusion of Social beliefs and Remote Sensing

Forests and arable land in Ghana face a significant threat due to the mechanisation and increase of illegal gold mining (galamsey). These directly affect local’s diets and nutrition, impacting communities that rely on forest resources and small-scale farming for sustenance. In regional systems, accurately predicting future outcomes is a crucial task, with applications ranging from environmental management to agriculture. Recent advances in participatory science and modelling have highlighted the potential of building collective models based on the knowledge and beliefs of local populations who interact with the system. Such approaches have provided more accurate estimations of future outcomes compared to traditional expert-driven methods. Under the HEURISTICS project, we are exploring the causal efficacy of local knowledge, beliefs, and attitudes in local communities' decisions on transition to agriculture, forest and Galamsey. Using unsupervised and supervised classification, different land uses and land covers (LULC) are classified from 2017 - 2023, including: Forest, Croplands, Settlements, Water, and Galamsey/Mining Sites. In addition to EO data, open-source data from OpenStreetMap are extracted, providing valuable information on roads, rivers, water streams, and administrative boundaries. To further enrich the data, machine reading models are employed to extract beliefs from articles, ranking them based on relevance to topics such as Galamsey, mining, cities, and settlements. Additionally, we leverage text data to map public sentiment towards mining activities. By analysing the origin and sentiment of sentences, we gain insights into how people perceive different areas and how these perceptions are connected to land use. Further analysis examines the factors influencing sentiment scores, including mining proximity, boundary effects, and authority influence. Grid-level sentiment maps reveal nuanced spatial patterns and highlight areas potentially impacted by mining. We also predict future mining trajectories by using machine learning models trained on historical  2017-2022 text and mining data that allows us to make a prediction for the 2023 year, and to identify key  factors correlated with the mining activities. An R-squared value of 0.94 was obtained, indicating that our approach explains 94% of the variance in mining proportion.  Keywords: Land use prediction, mining activity, Galamsey, sentiment analysis, remote sensing, Ghana.