Temperature and Precipitation Forecasts in Complex Terrain Using Planetary Boundary Layer Parameterization Ensembles

Turbulent exchanges of heat, moisture, and momentum between the Earth’s surface and atmosphere are parameterized in numerical weather prediction models by planetary boundary layer (PBL) schemes. These PBL schemes have been devised from experimental data over homogeneous terrain, yet are applied globally. A comparison of deterministic forecast performance using different PBL schemes has not been done in the complex terrain of western Canada.

We compared forecasts using eight different PBL schemes in the Weather Research and Forecasting model to one year of temperature and precipitation observations at 31 stations in British Columbia and Alberta. While the PBL scheme that produced the lowest mean absolute error (MAE) was location specific, the Mellor-Yamada-Janjić and Medium Range Forecast schemes had the lowest MAE for temperature and precipitation, respectively, when averaged over all stations.

Since it is difficult to know which PBL scheme produces the lowest MAE a priori, we tested the performance of an ensemble of PBL schemes. We evaluated three ensemble-member weighting schemes: equal weighting, weighting by the inverse of the ensemble-member root mean square error (RMSE), and multiple linear regression. For temperature, weighting the ensemble members by the inverse of their RMSE resulted in an ensemble forecast that matched the accuracy of the best-performing deterministic forecast. For precipitation, weighting all ensemble members equally resulted in the best ensemble forecast. Our results showed that multi-PBL scheme ensembles are not necessary when the most accurate PBL scheme is known beforehand, but using simple ensemble-member weighting schemes can produce forecasts with comparable accuracy as the best deterministic ensemble member.