Calibrating an All-Lands Wind Erosion Model for Rangelands

Abstract:

Aeolian processes play a fundamental role in structure and function of arid and semi-arid ecosystems. Despite this recognition, modeling approaches suitable for assessing impacts of rangeland management and environmental change on sediment transport rates over relevant spatial and temporal scales are poorly developed. For model estimates to provide value in this context, distributions of predicted sediment flux that encapsulate spatial, intra- and inter-annual variability are needed. Further, it is important to quantify and communicate transparent estimates of model uncertainty. In response, we parametrized the Aeolian EROsion (AERO) wind erosion and dust emission model for rangelands using a Generalized Likelihood Uncertainty Estimation framework (GLUE). GLUE is based on the concept of equifinality—many models can adequately describe an environmental process. As such, model estimates comprise a distribution rather than a single value, unknown model structural errors are implicitly accounted for, and uncertainty is quantified. Here, we present calibration results of the horizontal flux component of AERO. The model was calibrated using data from five sites across rangeland ecological states. Observations of wind speed, vegetation height, cover, and canopy gap length were input to the model to predict flux for 10,000 independently sampled parameter sets, and results compared to observations from 44 sediment transport observation periods (27 samplers per site * ~1-month collection period for each) to produce likelihoods of each model iteration. Selected rejection criteria resulted in 435 acceptable parameter sets and uncertainty was estimated for 90% prediction bounds using quantiles of ranked estimates from acceptable models. Results show good agreement for individual sampling periods across sites, with few observations falling outside of the 90% prediction bounds and overall accuracy of RMSE ≈ 10 (g m^-1 d^-1) when comparing mean predictions to observations. More importantly, combined distributions of flux estimates from all sample periods for a given site closely approximate the probability of observing a given flux at that site, suggesting this approach successfully encapsulates variability in transport rates and provides robust assessments for management.