Intercomparison of atmospheric forcing datasets and two PBL schemes for precipitation modelling over a coastal valley in northern British Columbia, Canada

Environmental modelling of remote areas requires dynamical downscaling of meteorological data to obtain precipitation values that could substitute for sparse in-situ observations. This study examined numerical simulations of precipitation over the Terrace-Kitimat Valley, an industrializing corridor in the Coast Mountains of northern British Columbia, Canada. Modelling uncertainty was explored for 1 year of output from the Weather Research and Forecasting model at 1-km grid spacing for three atmospheric forcing datasets and two planetary boundary layer (PBL) schemes. The observed total precipitation ranged from 1170 to 2380 mm and was often underestimated by more than 40% when using the North American Regional Reanalysis as atmospheric forcing data or the Mellor-Yamada-Nakanishi-Niino level 3 (MYNN3) parameterization as PBL scheme. Persistent low bias from model configurations using these configurations suggested that merely selecting an alternative atmospheric forcing dataset does not ameliorate systematic error occasioned by a poorly performing PBL parameterization. Hence, the choice of the PBL scheme and the meteorological dataset is important for spatial estimation of precipitation using WRF. Model output best corresponded with annual gauge measurements when simulations with the Mellor-Yamada-Janjic (MYJ) PBL scheme were forced with ERA5. The North American Mesoscale Analyses (NAM-ANL) however demonstrated better performance for monthly variation and high-intensity precipitation than ERA5. Using both datasets therefore may be valuable for calculations related to environmental change. With either NAM-ANL or ERA5 as atmospheric forcing data and MYJ as the PBL scheme, the uncertainty in annual simulated precipitation amount ranged between 38% overestimation and 21% underestimation of observational data.