<p>Integrating LEO satellite constellations with terrestrial networks is crucial for achieving seamless global coverage, ultra-low latency, and enhanced network reliability in future 5G/6G communications. This hybrid architecture enables ubiquitous connectivity, especially in remote and underserved areas. However, spectrum planning and interference management are still the major concerns in these integrated networks. To overcome these issues, this research proposes a Deep Reinforcement Learning-based spectrum planning and interference management in terrestrial and non-terrestrial LEO satellite constellations, named DRL-LEO. The proposed work entails three key processes delineated as cooperative clustering, intellectual spectrum planning and optimal interference management. Co-operative clustering is performed where both UE and LEO satellites are clustered by the Modified Location Weight-based Clustering Algorithm (MoLoClus). For the intellectual spectrum planning, the Twin Delayed Deep Deterministic Policy Gradient (TD3) Algorithm is incorporated, where the TD3 agents identify whether the spectrum is vacant or non-vacant based on the mother node (MN) aggregated information. Furthermore, the users are directed to the available spectrum with the utilization of CR, thus enhancing spectrum planning and spectrum sensing. Following that, the optimal interference management is implemented for amplifying Quality of Service(QoS), where the interference is optimally managed in two ways, thereby performing spectrum handoff and intellectual channel switching. Besides, through the adaptation of Hyped-up Hummingbird Optimization (H2O) Algorithm, the link of interest is optimally identified for accommodating efficacious channel switching.</p>