The agricultural sector faces a significant challenge: producing more food and generating higher revenue using less water. This challenge is exacerbated by the increasing scarcity of water resources due to climate change, population growth, and other factors, making optimal irrigation management crucial for sustainable agriculture. One of the primary tasks in this context is intra-seasonal irrigation scheduling under a limited seasonal water supply. This involves distributing a finite amount of water across multiple irrigation events throughout the growing season while considering the crop response to water stress at various growth stages. Effective management of this process using a deficit irrigation (DI) strategy can lead to improved water productivity and crop yields, thereby addressing the dual goals of food security and conservation of water resources in agriculture. This study aims to advance deep reinforcement learning (DRL) for DI systems and to benchmark a new deep reinforcement learning (DRL) approach against existing DRL strategies [1] for the closed-loop control of irrigation scheduling using the Aquacrop-OSPy model [2]. The evaluation is conducted under various conditions of water scarcity and climate uncertainty, incorporating detailed information about the state of the irrigation system and the climate environment. By considering these factors, the presentation provides a comprehensive assessment of the effectiveness of DRL in optimizing irrigation practices, particularly in scenarios characterized by limited water availability and changing climatic conditions.