Precision irrigation is necessary in enhancing water-use efficiency and crop productivity in the face of growing climatic variation and resource limitations. Conventional rule-based and standalone machine learning solutions would not be able to embrace the short-term nonlinear interrelations of environmental factors as well as long-term time-related dependencies in irrigation demand. In an attempt to overcome these limitations, this paper introduces a hybrid LightGBM-Autoformer model to predict-based precision irrigation control. LightGBM is also used in the proposed model to capture nonlinear interactions between soil moisture, temperature, and humidity among other agro-meteorological variables, and Autoformer is used to predict future irrigation needs due to its effective ability to learn long-term temporal dynamics in time-series data. These models have been combined to allow proactive and adaptive scheduling of irrigation instead of a reactive response. The experimental evidence shows that the suggested hybrid framework is more effective than the current baselines of machine learning and deep learning in terms of prediction accuracy, stability, and computational efficiency. The results show that the framework can considerably enhance the water-use optimization and decision reliability, hence it can be deployed to resource-constrained edge and TinyML-enabled IoT platforms. This research brings out the practicality of the enactment of boosting-based regression and transformer-based forecasting models in sustaining intelligent water management in agriculture.