Scale- and Vegetation-Dependent Energy Flux Biases in CoLM2024 and ECLand: A PLUMBER2 Evaluation

Simulation of the surface energy balance (SEB) is essential for land–atmosphere coupling and weather–climate prediction, yet land surface models remain uncertain in turbulent and ground heat fluxes. We evaluated CoLM2024 with Land Cover Type (LCT) and Plant Community (PC) schemes and ECLand v1.0 against energy-balance-corrected observations from 80 PLUMBER2 towers spanning 11 land cover types. Observations and simulations were decomposed at 30-min, daily, and monthly scales. All experiments reproduced net radiation well, whereas ground heat flux was poorly simulated over forests and wetlands because of excessive daytime amplitude, indicating limitations in canopy–soil heat partitioning and soil heat storage. ECLand achieved the best latent heat flux performance through smaller systematic errors. PC reduced unsystematic errors, but this benefit was offset by a large negative growing-season bias. Sensible heat flux performance depended on vegetation: PC performed best over evergreen needleleaf and mixed forests, while ECLand was superior over broadleaf forests and had the lowest unsystematic errors, although its implicit coupling damped variability. Performance was scale dependent: ECLand was favored at 30 min, PC was competitive for daily forest sensible heat correlations, and no consistent ranking emerged monthly. Complex canopy schemes therefore require robust formulations, improved heat-storage processes, and better parameter calibration.