Augmenting The Weather: A Hybrid Counterfactual-SMOTE Algorithm for Improving Crop Growth Prediction When Climate Changes

AI struggles with climate extremes because it learns from historical patterns that don't include rare, extreme weather events. The paper identifies a fundamental problem: machine learning models trained on past weather data perform poorly when predicting unprecedented climate events, since these outlier situations are underrepresented in historical datasets.

The proposed CFA-SMOTE method artificially creates synthetic examples of extreme weather scenarios to train AI models better. By combining two existing techniques (counterfactual reasoning from explainable AI and SMOTE, a class-imbalance method), the researchers generated synthetic data representing climate outlier events, which improved prediction accuracy for crop growth on Irish dairy farms during droughts.

This approach addresses a critical gap in responsible AI for climate adaptation: making AI systems robust to unprecedented conditions. Rather than relying solely on historical data, the method helps ensure AI tools can better handle the "new normal" of climate change, which is essential for helping farmers and agricultural sectors adapt to extreme weather.

The research demonstrates practical application in agriculture, a sector highly vulnerable to climate disruption. Testing on grass growth prediction for Irish dairy farms shows how this technique could help farmers make better decisions during droughts and forage crises—real-world scenarios where AI predictions currently fail.