Crop yield forecasting is a fundamental part of current precision agriculture which enables sustainable farming practice and optimal resources scheduling. Combining IoT with deep learning can improve predictive performance by benefiting from the realtime data gathering and the power of computational models. In this study, an IoT-based crop production forecasting framework is introduced which uses the LEO, during the progress of the system to effectively route the data, so that the energy consumption will be minimum and the data transmitted is precise. The collected data is pre-processed and features are extracted by a Temporal Convolutional Network (TCN) to capture longrange dependencies of agricultural-related data. These representations are passed to a hybrid BiLSTM-WANN network which integrates a BiLSTM to capture the time dependency followed by a WANN network that optimizes the model structure without weight updates. What is more, we also propose to LEO to update weight parameters in order to reduce prediction error and increase accuracy. This implication results in an efficient, scalable, and high-performance architecture for crop yield prediction. The performance of the proposed method is MSE=0.0082 and MAE=0.003 less than that of the existing methods. Through the combination of optimized routing, deep learning, and metaheuristic optimization, the system improves the agricultural decision-making process, resource use optimization, and environmentally friendly farming, positioning itself as an asset in precision agriculture.

The increasing complexity of smart city ecosystems and the demand for real-time urban governance present challenges in privacy preservation, communication overhead, scalability, transparency, and adaptive decision-making. Existing frameworks often rely on centralized architectures and investigate digital twins, federated learning, explainable artificial intelligence, and agentic intelligence independently, limiting coordination and sustainability. This study proposes an Edge-Enabled Digital Twin-Based Federated Explainable Agentic Artificial Intelligence framework for real-time sustainable smart city governance. The proposed framework integrates edge intelligence, digital twin synchronization, federated learning, explainable decision-making, and agentic reasoning within a unified architecture to enable collaborative and adaptive urban intelligence. Representative datasets from transportation, energy management, environmental monitoring, healthcare, and public safety are used for evaluation in a distributed simulation environment. Experimental results demonstrate that the proposed framework outperforms conventional intelligent approaches by improving decision accuracy, reducing communication overhead, and enhancing transparency and responsiveness. Overall, the proposed framework provides a scalable, privacy-aware, and trustworthy foundation for sustainable smart city governance through transparent, adaptive, and intelligent decision support.