2026-10-14T20:33:41Z https://www.scidb.cn/oai

10.57760/sciencedb.012az 2026-09-14T09:40:14Z

2026-09-14 Data package for "A differentiable PyTorch implementation of the OpenIFS land-surface physics": initial states, observation subsets, Fortran reference outputs, and archived experiment results doi:10.57760/sciencedb.012az en This data package accompanies the manuscript "A differentiable PyTorch implementation of the OpenIFS land-surface physics" (submitted to Geoscientific Model Development) and backs every reported number and figure. It contains: the shared initial states of all experiments (the ERA5-Land 2005 state mapped onto the 97,709 CMFD land columns, and the resulting five-year spun-up equilibrium state); the 2018 MODIS subset (MOD11C1 land surface temperature and MOD10C1 snow cover) sampled at the CMFD column centres; the offline Fortran reference outputs used in all equivalence comparisons (the 30-day controlled experiment, the one-year 2018 integration, and the per-year terminal states of the 2009–2018 chained integration); the year-long MODIS observation-operator products of the uncalibrated model and of the calibrated two-timescale surface-layer closure (SURF-Opt); archived result records of all experiments (metrics, training histories, logs, figures); the manuscript sources and compiled PDF; and per-experiment manifests with SHA-256 source hashes together with the documented reproduction chain. The model implementation source code and run scripts are available from the corresponding author upon request and will be released publicly upon publication; the raw driving datasets (CMFD, ERA5-Land, MODIS) are public and cited by reference. and surface model; OpenIFS SURF; PyTorch; automatic differentiation; land surface temperature; MODIS; CMFD; CUDA graph; land data assimilation Yang Jinhui PUBLIC https://mit-license.org dataset Science Data Bank