Climate models (CM) are used to evaluate the impact of climate change on the of floods and strong precipitation events. However, these numerical have difficulties representing precipitation events accurately, due to limited spatial resolution when simulating multi-scale dynamics the atmosphere. To improve the prediction of high resolution precipitation apply a Deep Learning (DL) approach using an input of CM simulations of the fields (weather variables) that are more predictable than local. To this end, we present TRU-NET (Temporal Recurrent U-Net), an-decoder model featuring a novel 2D cross attention mechanism between convolutional-recurrent layers to effectively model multi-scale-temporal weather processes. We use a conditional-continuous loss to capture the zero-skewed %extreme event patterns of rainfall. show that our model consistently attains lower RMSE and MAE scores a DL model prevalent in short term precipitation prediction and improves the rainfall predictions of a state-of-the-art dynamical weather model., by evaluating the performance of our model under various, training testing, data formulation strategies, we show that there is enough data for deep learning approach to output robust, high-quality results across and varying regions.