We evaluate the extent to which distinct semi-arid environments and model parameterizations affect the behavior of ‘physically similar' land-surface model parameters. We draw preliminary conclusions about their temporal and spatial variability.

We drive the NOAH and CLM land-surface models offline and perform several multi-criteria sensitivity analysis and parameter estimations using 2+ years of quality controlled flux tower data from a grass site and a shrub site in Arizona (USDA Lucky Hills and Kendall) and two similar sites in New Mexico (Sevilleta LTER).

For each model we cross-validate estimated optimal parameter sets and their sensitivity by comparing parameters obtained at one site against those obtained at other sites and those resulting from simultaneous optimization at all sites. This single model analysis of parameter transferability allows exploring site-to-site differences for two dominant biomes: shrub and grass due to hydroclimatic conditions.

We then evaluate the consistency and transferability of parameters with similar physical meanings across models and investigate the effect specific model parameterizations of land-surface processes affect estimated parameter values and their sensitivity.

The results presented follow the protocol of the second stage PILPS Semi-Arid Experiment for rigorous evaluation of land-surface models performance in semi-arid environments.

Addressing the issue of parameter transferability between sites with different vegetation covers, located a few kilometers apart and the transferability of parameters between sites with similar hydroclimatic regime but hundreds of kilometers apart also has implications for ungauged catchments, regionalization efforts, etc.