Climate change has created uncertainty in rainfall hence becoming a major problem in agriculture especially in places dependent on rainfall to irrigate farms. This paper suggests deep-learning-based IoT framework to predict rainfall real time and plan irrigation adaptively in precision agriculture. The system concept is a model of the IoT sensor networks, which are used to create the agro-meteorological data, such as temperature, humidity, atmospheric pressure, and soil moisture. This data is utilized in training a Long Short-Term Memory (LSTM) model, which was chosen due to successful performance in modeling nonlinear, time-series weather patterns. The forecasted results of rainfall are also incorporated in a simulation grounded adaptive logic of irrigation which shows how such forecasts can inform water efficient method of irrigation. The work presents a contrast with the currently available works as it aims at the conceptual design, simulation of the proposed model as well as its performance evaluation which are not stated in the existing works to the greatest extent. The output of experimentations on given benchmark agricultural datasets demonstrates that LSTM-based prediction model with regard to accuracy and responsiveness surpasses the traditionally used statistical and machine learning methods. This framework provides a blueprint of a scalable, smart solution that can be developed in the future about climate-resilient agricultural support systems with the possibilities of AI and the IoT.