Human-Guided Causal Hypothesis Testing for Remote Sensing Anomaly Detection

Remote sensing anomaly detection models often succeed at recognizing anomalous patterns but struggle to provide causal and human-interpretable explanations for why an environmental change is considered anomalous. Human analysts frequently reason through causal hypothesis testing by proposing candidate causes, evaluating supporting evidence, and selecting explanations that best fit observations while remaining temporally and spatially coherent. We present CogChain, a cognitively inspired neurosymbolic reasoning layer that augments neural anomaly detectors with structured causal hypothesis testing over short causal chains. CogChain combines neural evidence extraction with probabilistic inference regularized by human-inspired priors including temporal causality, spatial contiguity, and simplicity. Experiments on remote sensing anomaly detection settings using EuroSAT and SEN12MS demonstrate improved anomaly detection performance and interpretable chain-structured explanations compared to neural-only baselines. Our findings suggest that integrating lightweight causal reasoning into scientific imaging workflows may improve interpretability and robustness for AI-assisted environmental monitoring.