Forecast Skill of Varying WRF Resolutions and Physics Parameterization Combinations over the Finger Lakes and Long Island with Statistical Postprocessing

Short-range forecast skill of varying WRF resolutions and physics parameterization combinations is analyzed over the Finger Lakes, New York (FL) and Long Island, New York (LI) with and without statistical post-processing. Both areas possess Network for Environment and Weather Applications (NEWA) station data, which provide desired observations for microclimates in these regions and allow the forecasts to be compared relative to the agricultural decisions they influence. WRF is initialized with GFS 0.5-degree data, and set up to run for four days at 00z every day during the growing season (March-October) in 2009-2018. An ensemble WRF forecast is also created, with ensemble members each having different combinations of physics parameterizations. 27km, 9km, and 3km WRF domains are utilized, with forecast skill calculated directly from the WRF output. Furthermore, a nonhomogenous Gaussian regression is applied to all three of the WRF ensemble domains, with a unique training period generated for each station based on trial runs. Station data from NEWA are used to post-process and assess the forecast skill of the model output. Model skill is also compared with that of forecasts from the National Digital Forecast Database.

Initial results from single domain WRF runs (20km) demonstrate a marked increase in skill from statistically post-processed ensemble forecasts over raw output ensemble forecasts. The methods will be applied to the aforementioned domains to investigate whether post-processed lower resolution WRF significantly outperforms higher resolution WRF in the FL and LI microclimate regions in terms of the crop management decision to which the forecasts are applied.