EMS2017 – An Investigation of Systematic Errors in Solar Radiation for Reanalysis Datasets

I attended EMS2017 in DCU, Dublin in September 2017. I presented a poster titled “An Investigation of Systematic Errors in Solar Radiation for Reanalysis Datasets”. This included the accuracy of reanalyses to estimate shortwave radiation (SW) compared to station observations, using a selection of standard skill scores. It also shows initial results of a systematic link between prevailing cloud structures (frontal cloud or convective cloud) and SW errors, as seen in satellite imagery. In a limited sample of cases, for MERRA2, convective clouds are linked to large negative errors and frontal clouds are linked to large positive errors. The final part of the poster was to show the improvement a simple adaptive post processing method (linear least squares) can provide to the reanalysis skill scores.




Journal Club: 04-04-2018

Article Title: Using weather ensemble prediction in electricity forecasting, Taylor and Buizza (2013).

The authors compare different methods of including weather into an electricity demand forecasting model. They compare the traditional method (inputting the mean of an ensemble of weather forecasts), to that of applying each weather forecast from the ensemble to the demand model, and then taking the mean of the scenarios. They show that the second method (x) performs better than the traditional method (o) for lead times of 1-10days (see Figure 3 below).

The difference in accuracy of the 2 methods increases with lead time. It is concluded that the reason for the improved performance is due to the fact that the demand is a non-linear function of weather. Further, they show that excluding weather variables from the electricity demand forecast model reduces the accuracy substantially.

The demand model is a regression model based on the UK National Grid. Weather related components depend on effective temperature, cooling power of the wind and effective illumination (replaced by cloud cover here). The model is weighted to reflect different population densities across 5 English cities.

The authors also present different methods of estimating the demand forecast error and the demand prediction intervals. From this they conclude that including weather ensemble information can improve these estimates compared to traditional volatility forecasting methods.

Applications to our research:

  • Implement this demand model for Ireland.
  • Use this model to identify atmospheric spatial patterns associated with certain demand profiles.
  • Apply Shortwave Radiation data to model instead of cloud cover.
  • Potentially use the model to link demand profiles to climate patterns (eg. NAO).
  • First check if there are improved versions of this model.



WRF driven by GFS Ensemble data on ORR2

GFS Ensemble data are available to download from here:

  • Grid GENS 3
  • Start Date = End Date. Select files. I’m assuming that _00 is the “control” forecast, and _01 to _20 are the perturbed ensemble members.
  • When your data are ready to download, you will get an email notification. Click on the “Web Download” link. Each ensemble forecast member will have a link that looks something like this:
    gens_3_2017101200_00.g2.tar 03-Apr-2018 07:06 1.6G
    Right-click on the file and copy the link address. Now ssh into the ORR2 computer. Create a folder for this data, like GRIB/GENS/GENS00, and use wget with your copied link to download the data file.
  • Unpack your forecast data file using the tar command, e.g.:
    tar -xf gens_3_2017101200_12.g2.tar
  • You should now have lots of .grb2 files in that directory. You’re ready to run WRF.

The steps to run WRF driven by GENS are similar to those to run WRF with GFS, so follow this post: A few differences to note:

  • link Vtable to ungrib/Variable_Tables/Vtable.GFSENS
  • ./link_grib.csh to both the gens-a and gens-b files.
  • In namelist.input, use: num_metgrid_levels = 27,

If all goes well, you should be able to run your WRF forecast. WRF doesn’t output MSLP, but you can use a code like this to calculate and plot it: