Spatially-uneven observations cause overestimation of global soil respiration

This repository (Part I: Code, Core Data, and Prediction Maps Part 1) provides the complete reproducible workflow and core datasets for estimating global soil respiration while correcting for spatial sampling bias. The workflow includes four sequential scripts: (1) 00_feature_extraction.py extracts environmental raster values to observation points, converting the raw input srdb-data-v2.csv into the enriched srdb-data-with-raster-values.csv; (2) 01_pruning.py applies a Voronoi-PCA algorithm to generate optimized subsets, outputting optimal_subset.zip and an internal cache srdb_raster_extracted.csv; (3) 02_validation.py performs 10-fold cross-validation to evaluate model performance; and (4) 03_prediction.py generates the final global soil respiration maps (1982–2022). This Part I record contains the source code, core tabular data, and the first four segments of the global prediction dataset (rs_global_dataset_v1.7z.001 to rs_global_dataset_v1.7z.004). To reconstruct the complete prediction maps, users must also download Part II (DOI: [10.5281/zenodo.18948381]), which contains the remaining segments (rs_global_dataset_v1.7z.005 to .009) and the cross-validation uncertainty estimates (cv_global_dataset_v1.7z). After downloading all nine .7z parts from both records into the same directory, extract the first file (rs_global_dataset_v1.7z.001) to automatically reconstruct the full global soil respiration dataset. Both Zenodo records should be cited together to ensure full reproducibility of the study.