Weather forecasting is essential for many industries, but especially for agriculture, where precise forecasts have a big impact on financial results. Because weather patterns are dynamic, traditional statistical methods frequently fall short of providing accurate forecasts. Recent advancements in Long Short-Term Memory (LSTM) networks offer encouraging solutions. However, challenges such as the integration of additional weather parameters, handling data noise, and improving model accuracy remain significant hurdles. This paper proposes a time series-based weather forecasting system using LSTM to address these challenges. By incorporating advanced LSTM architectures, including Stacked and Bidirectional LSTM, and preprocessing techniques to enhance data quality, the model aims to enhance accuracy. The experimental results demonstrate that the proposed model outperforms traditional approaches, achieving lower error rates and better generalization to unseen data. The ultimate objective is to give farmers and other stakeholders quick and accurate weather forecasts so they can make knowledgeable choices and take preventative actions to lower risks and minimize losses in the agriculture sector.