LandFormer Dataset: Meteorological Forcing, VIC Model Parameters, and Simulated Hydrological States for Continental-Scale Modeling

Description This dataset provides the processed inputs and targets used in the development and evaluation of LandFormer, a Swin Transformer–based surrogate model for continental-scale land surface modeling over the Indian subcontinent. The dataset consists of three main components: 1. Meteorological Forcing Data Gridded daily meteorological forcing derived from ERA5 and regridded to a uniform spatial resolution using bilinear interpolation. The variables include: Downward longwave radiation (lwdown) Downward shortwave radiation (swdown) Precipitation (prcp) Surface pressure (sp) Maximum temperature (tasmax) Minimum temperature (tasmin) Vapor pressure (vp) Wind speed (wind) The forcing data span multiple decades and are structured as (time, latitude, longitude) grids covering the Indian subcontinent. 2. VIC Model Parameter Data This component contains the full set of Variable Infiltration Capacity (VIC) model parameters at high spatial resolution. Parameters include: Soil and hydraulic parameters: infilt, Ds, Dsmax, Ksats (multiple layers) Soil layer depths: d1, d2, d3 Exponential parameters: expt1, expt2, expt3. etc All parameters are organized as spatial grids (latitude, longitude) or (nfiles, latitude, longitude) depending on parameter structure. 3. VIC Simulated Outputs Outputs generated using the VIC model (version 4.2d), which serve as target variables for training and evaluation of the surrogate model: Streamflow (Q) Evapotranspiration (ET) Soil moisture (S) These outputs are provided as spatiotemporal grids aligned with the forcing data. Data Characteristics Spatial coverage: Indian subcontinent Spatial resolution: ~0.0625° Temporal resolution: Daily Format: Zarr