Global XCH₄ Data Product (2024–2025) from TROPOMI and GOSAT using Transfer Learning

Global XCH₄ Dataset (2024–2025) is generated using a multi-satellite transfer learning framework, covering the period from 1 January 2024 to 31 December 2025. The retrieval algorithm employs a probabilistic multilayer perceptron neural network pretrained on TROPOMI observations and fine-tuned with high-accuracy GOSAT retrievals. The data product is stored in Parquet format with native TROPOMI spatial resolution and is organized as monthly files named by year–month (YYYYMM.parquet, e.g., 202401.parquet). Each file contains seven variables: time_sequence (orbit sensing start and end times inherited from the original TROPOMI filename, e.g., 20240101T074458_20240101T092629), longitude, latitude, scanline, and pixel (extracted from original TROPOMI products), together with methane_mixing_ratio (XCH₄ in ppb) and uncertainty (pixel-level uncertainty estimates) provided by this study. The geographical domain spans 60°S–90°N and 180°W–180°E (global terrestrial regions, excluding Antarctica). For detailed access information, contact the corresponding author (tao.ren@sjtu.edu.cn).