Geographia Technica, Vol 21(2), Special Issue: Artificial Intelligence Applications in Geography, 2026, pp. 119-133
HYBRID DEEP LEARNING AND REINFORCEMENT FRAMEWORK FOR PHYSICALLY CONSISTENT PRECIPITATION CONTROL ALONG THE MOROCCAN-IBERIAN COASTS
Imad KATIBA 
, Hanae BELMAJDOUB , Khalid MINAOUI
ABSTRACT: This study introduces a hybrid framework that integrates deep learning and reinforcement learning to optimize regional precipitation along the Moroccan and Iberian coasts of the North-Eastern Atlantic. A Bidirectional Long Short-Term Memory (BiLSTM) and Multi-Layer Perceptron (MLP) network was first trained to forecast monthly rainfall from multi-variable climate sequences combining key atmospheric and oceanic variables. The predictions were then coupled with a Twin Delayed Deep Deterministic Policy Gradient (TD3) controller capable of performing small, physically coherent adjustments to selected variables such as sea-level pressure, specific humidity, and near-surface winds. Through adaptive interactions with the predictive environment, the control agent learned strategies that enhance rainfall while maintaining atmospheric stability. The resulting framework demonstrates consistent improvements in simulated precipitation and reveals spatially coherent dominance patterns of sea-level pressure and humidity in rainfall modulation. The reinforcement-based control explicitly enforces stability and energetic constraints, ensuring physically admissible atmospheric responses. Overall, the proposed hybrid AI approach provides a physically interpretable foundation for regional water management and constitutes a decision-support framework for diagnosing precipitation sensitivity rather than an operational weather modification system. It can be extended to drought-prone or broader climatic domains to support future resilience and hydrological planning.
Keywords: Deep learning; Reinforcement learning; BiLSTM; TD3; Climate control; Precipitation optimization; Moroccan–Iberian coasts; North-Eastern Atlantic; Atmospheric dynamics; Water resource management.
Full article here