Global Total Projected Leaf Area Index Subset from the Community Land Model Version 5 (CLM5) Perturbed Parameter Ensemble Members, 1850-2015

Description

This dataset contains a subset of total projected leaf area index (LAI) simulations derived from a 500-member Perturbed Parameter Ensemble (PPE) generated using the Community Land Model version 5 (CLM5). The data spans from February 1850 to January 2015 and covers 400 global sparse grid cells at a 4° × 5° resolution. Each ensemble member reflects a unique configuration of 32 perturbed parameters relevant to stomatal conductance, photosynthesis, hydrology, biogeochemistry, and phenology.

This LAI subset was produced by the Climate and Global Dynamics (CGD) Laboratory at the National Center for Atmospheric Research (NCAR) using GSWP3 meteorological forcing. The ensemble was constructed using Latin Hypercube Sampling to enable systematic exploration of parametric uncertainty in CLM5, with the goal of supporting sensitivity analysis, emulation, and machine learning-based surrogate modeling.

These LAI outputs were used to train an Evidential Deep Neural Network (EDNN) emulator developed under the EDNN LAI project, which aims to reproduce LAI dynamics while quantifying both aleatoric and epistemic uncertainty. The subset facilitates efficient emulation and enables uncertainty-aware exploration of vegetation-climate interactions.

Dataset Characteristics

• Temporal Coverage: 1850-02 to 2015-01

• Spatial Coverage: Global (subset of 400 sparse grid cells)

• Variables:

  • TLAI: Total Projected Leaf Area Index (m²/m²)

• Dimensions:

  • member: 500 ensemble members

  • time: Monthly time steps (1980 months total)

  • gridcell: 400 global sparse grid cells

• Units: m²/m²

• Cell Method: Time: mean

• Coordinate System: No-leap calendar (cfTime)

• Data Format: NetCDF, compatible with xarray and dask

Usage Notes

This dataset is designed for surrogate modeling applications, especially in contexts requiring uncertainty quantification or parameter sensitivity analysis. It has been used to train deep learning emulators for emulating CLM5 outputs at reduced computational cost. Users should account for the no-leap calendar and the PPE configuration when analyzing long-term trends or aggregating across ensemble members.