Dataset for "Skill-Equivalent Deep Learning Models Learn Physically Distinct Attribution Patterns in Precipitation Downscaling"

This dataset contains model outputs, Integrated Gradients (IG) attribution data, ablation experiment results, and pre-processed statistical summaries used in the study "Skill-Equivalent Deep Learning Models Learn Physically Distinct Attribution Patterns in Precipitation Downscaling". The study investigates the explainability and physical interpretability of three deep learning architectures (FSRCNN, SRDRN, and U-Net) trained for statistical downscaling of daily precipitation over India.本数据集包含用于研究《技能等价深度学习模型在降水降尺度任务中学习到物理差异化归因模式》（Skill-Equivalent Deep Learning Models Learn Physically Distinct Attribution Patterns in Precipitation Downscaling）的模型输出、集成梯度（Integrated Gradients，IG）归因数据、消融实验结果以及预处理后的统计汇总信息。该研究针对为印度区域日降水统计降尺度任务训练的三种深度学习架构FSRCNN、SRDRN与U-Net，探究了其可解释性与物理可解释性。