Impact of Temperature on Rainfall Erosivity on the Loess Plateau: An Innovative EMA-Pix2PixHD Framework

This dataset contains the complete research data from the EMA-Pix2PixHD-based prediction of rainfall erosivity (R) over the Loess Plateau. It includes: (1) the trained generator models G1 and G2; (2) generated R images on the training set; (3) monthly R remote sensing images projected under SSP245 and SSP585 scenarios for the years 2040, 2070, and 2100; (4) attention heatmaps from the EMA module for interpretability analysis, revealing model focus on key climate-sensitive regions; and (5) evaluation outputs, including CSV files of MAE and RMSE for each test image, and pixel-wise relative error (RE) maps. All R images are inverse-normalized to physical units (MJ·mm·hm⁻²·h⁻¹), and RE values are expressed as percentages. This dataset supports research on soil erosion risk assessment under climate change, geoscientific applications of deep learning, and the reproducibility and extension of image-to-image translation models.