The uncertainty in model forecasts is commonly estimated from the variability within a ensemble (set) of multiple model forecasts, in which each forecast is adjusted, or perturbed, to account for potential sources of error in the model forecasts. Currently, the ensembles used to estimate the uncertainty in weather forecast models underestimate the uncertainty at and near the land surface, because the ensemble is not perturbed to account for uncertainty in the land model.

In a new Physical Sciences Laboratory (PSL) study to be published in the Journal of Hydrometeorology, a suite of experiments tested different methods for perturbing ensemble forecasts to account for land model uncertainty in NOAA's Numerical Weather Prediction (NWP) model. The PSL researcher found that the most appropriate method was to perturb a selection of the model parameters used in the model equations for the exchange of water and energy between the land and atmosphere. This approach enhanced the ensemble variability at and near the land, so that the resulting ensemble-based forecast uncertainty estimates agreed well with independent forecast uncertainty estimates.

Data assimilation improves the initial conditions from which forecasts are run, by updating the model initial conditions with a weighted estimate of modeled and observed values, with the weights depending on the relative uncertainty in the model and observations. By improving the ensemble-based estimates of land forecast uncertainty, the above work will enable development of an ensemble-based land data assimilation system to introduce land observations into NOAA's NWP systems.

In addition to improving the uncertainty estimates of land forecasts, perturbing the land/atmosphere exchange parameters also generated ensembles with a realistic relationship between forecast errors in the land and the atmosphere. For coupled data assimilation, in which observations of one model component (the land) are used to update another model component (the atmosphere), this relationship between forecast errors in each component is central to the success of the data assimilation.

By introducing a method to account for land model uncertainty in NOAA's global NWP ensembles, this work will improve the forecast uncertainty estimates calculated from those ensembles. These improved uncertainty estimates are useful for probabilistic forecasts of variables influenced by the land surface (including precipitation and temperature). These estimates will also help PSL researchers develop an ensemble-based coupled land/atmosphere data assimilation system for NOAA's global NWP system, in which land states are updated from observations of atmospheric variables (and vice-versa).

Draper, C. S. (2021): Accounting for land model error in numerical weather prediction ensemble systems: toward ensemble-based coupled land/atmosphere data assimilation. J. Hydrometeorol., https://doi.org/10.1175/JHM-D-21-0016.1.

Posted: July 19, 2021