Geospatial Causal Principle and Causal Discovery for Geospatial Effects

The rise of geographic big data and the advancements in deep learning have brought new opportunities to solve challenges in geospatial analysis. Current methods are rooted in statistical correlations. However, the deceptive and misleading nature of correlation patterns, coupled with inherent biases in the data, make it difficult for those methods to obtain reliable analysis results. And the source of that difficulty is the fact that correlation in data does not imply causation. Serving as the key to unveiling unknown facts and discovering general underlying patterns, causation is critical to geographical research. Given the inherent complexity of earth system and that geographic researches are mainly based on observation into earth, it is impractical to conduct randomized double-blind experiments and inappropriate to directly apply the philosophical definition of causation in geographical researches. In light of that, this paper aims to provide fundamental principles and methodological support for geospatial causal analysis. On the basis of assumptions about geospatial causation, the principle of geospatial causation is proposed. Furthermore, the relationship between this principle and the spatial effects in geospatial analysis is elucidated, major pathways for causal discovery under the consideration of each spatial effect are also presented.