Evaluation of water, snow, energy, and carbon dynamics in kilometer-scale E3SM Land Model (ELMv2) across the Contiguous United States

Abstract. Kilometer-scale (k-scale) land surface modeling is increasingly recognized as essential for accurately simulating terrestrial water, energy, and carbon cycles. In this study, we evaluate and improve the performance of the E3SM Land Model version 2 (ELMv2), the land component of E3SM version 2, configured at 1 km resolution across the Contiguous United States (CONUS). Offline ELMv2 simulations incorporating a newly developed 1 km surface dataset and improved parameterizations for sub-grid topographic effects on solar radiation and pluvial inundation were performed for 2001–2019, driven by NLDAS2 atmospheric forcing. Two uncertain model parameters, the subsurface drainage parameter (f_drain) and the pluvial flood dynamics parameter (f_c), were sequentially calibrated against benchmark water table depth and surface water dynamic datasets, leading to substantial improvements for both variables relative to the default configuration. Additional ten independent datasets were used to evaluate ELM-simulated water, snow, near-surface energy, and carbon cycle dynamics. The calibrated ELM captured the broad spatial and seasonal patterns of soil moisture, runoff, snow water equivalent (SWE), latent heat flux, land surface temperature, and gross primary productivity (GPP) reasonably well, though systematic biases remained, including a wet bias in soil moisture over the Midwest and Northeast, underestimated and muted-amplitude runoff with a shifted seasonal peak, underprediction of SWE across the mountainous West, an overestimation of summertime latent heat flux over the south-central CONUS, and an overprediction of GPP over the central and eastern domain. Relative to the simulation with default f_drain and f_c values, calibration had minimal impact on model performance against the ten independent benchmark datasets. Upscaling the 1 km simulation to 12 km resulted in substantial decrease in the information loss metric (34–44% on average) in net radiation, latent heat flux, sensible heat flux, and near-surface soil moisture, highlighting the value of k-scale land surface modeling for resolving fine-scale terrestrial processes. These results demonstrate the value of k-scale ELM simulations for resolving fine-scale spatial heterogeneity in water, energy, and carbon fluxes, while also highlighting the continued need for improved model physics, calibration strategies, and higher-fidelity input datasets to further reduce systematic biases as ELM moves toward global k-scale application.