I presented a talk titled “WRF planetary boundary layer schemes for wind forecasting” in the “Mesoscale” session as part of Theme 1: “Wind resource, turbulence and wakes” at WESC in UCC. The other talks in the session included the following.
- Dr. Martin Doerenkaemper presented work involved in the production of the New European Wind Atlas (NEWA). He outlined computing hardware challenges of the project, WRF setup testing and workflow planning.
- Dr. Dries Allaerts presented on assimilation methods involving mesoscale-to-microscale coupling.
- Prof S.C. Pryor presented an analysis of 2 different wind farm parameterization, Fitch and EWP, and varying model resolution. These were run for an area over Iowa, USA. It was found that EWP produced 1.5-2.1% higher capacity factors compared to Fitch, which could impact wind farm planning. Further investigations needed to determine which scheme is closer to reality.
- Erkan Yilmaz presented work looking at WRF skill at forecasting wind in complex terrain in Central Turkey. Analysed the pros and cons of nearest point grid selection and bi-linear interpolation for better forecast results.
I presented a talk titled “An investigation of WRF PBL schemes for renewable energy forecasting in Ireland” in the “Forecasting for power system applications: wind models” at ICEM2019 in DTU.
- Sue Ellen Haupt presented about work on mesoscale-to-microscale coupling at NCAR for wind energy applications. Discussed the issue of “terra incognita”, the area between O(1km) and O(100m) between mesoscale NWP and LES. Said that upper limit of terra incognita is boundary layer height.
- James Wilzcak presented work as part of WFIP2, discussing the observations collected as part of the field campaign and how these led to model development within WRF as a result, e.g. looking at cold pools leading to sustained weak winds east of mountains in Washington state. WRF repository for their updated to MYNN on github.
- Sven-Erik Gryning presented work using WRF to downscale GFS forecast data to predict wind ramping at different forecast horizons at FINO3 platform. Forecasts were evaluated using correlation coefficient and comparison of histograms. At 10 min lead time, WRF performed poorly, becomes skillful at 4 hours lead time before reducing for longer lead times. Results shown for wind speed and wind direction.
- Jared Lee presented work on a project for renewable energy forecasting in Kuwait using WRF-Solar-Wind and DiCast (a multi-model forecast system using MOS to determine adaptive weighting of different models using a 90 day training period). Also using a “cubist” machine learning method.