Welcome to VAR_BLENDING documentation

VAR_BLENDING package provides a Variational based method to blend different NWP models (e.g., a global model and a limited area model) on a spectral space.

• Release: 0.0.1
• Date: 2018-12-28

This package is developed by New Zealand Meteorological Service, and to work with Amazon Web Service Storage Server S3

Questions go to: sijin.zhang@metservice.com

There are a few assumptions of using this method:

• a global model (e.g., GFS or ECMWF) is able to provide better large scale weather patterns than a limited area model (e.g., high resolution WRF)
• the truth of atmosphere lies between the global model and limited area model
• there are no correlations among the fields to be blended
• a global model and a limited area model contains similar amount of energies for the large scale patterns (e.g., where the wavenumber is bigger than a certain value)
• the imbalances among fields after the blending analysis can be solved by the model itself

About this software

• license
• help

Methodology

• mathematical ideas

Sample workflow

• Step 1: preparing model errors
  • STEP 1.1 Download (or prepare) the model inputs
    • 1.1.1 if download model inputs from AWS S3
    • 1.1.2 if copy model inputs from somewhere else
    • 1.1.3 check if all the required data are presented
  • STEP 1.2 Determine the model error estimation criteria
    • 1.2.1 It would be good if the background does not have significant large scale forcing:
    • 1.2.2 It would be good if the background is mostly dry:
    • 1.2.3 Determine if the blending can be continued
  • STEP 1.3 extract the model data
    • 1.3.1 Read global model forecasts and LAM forecasts
    • 1.3.2 Read global model analysis data
  • STEP 1.4 convert model data using FFT
    • 1.4.1 convert 2D grid data using 2D FFT:
    • 1.4.2 create the power from the grid data:
    • 1.4.3 return standard powers and FFT coeffs
    • 1.4.4 create the Welch power vertical profile from the grid data:
  • STEP 1.5 calculate model errors
    • STEP 1.5.1 Running inverse FFT for global/LAM forecasts and global analysis for all required wavenumbers:
    • STEP 1.5.2 Obtain the user defined error ratio
    • STEP 1.5.3 Obtain the model mean values over levels
    • STEP 1.5.4 Obtain the model errors
• Step 2: running Variational blending
  • STEP 2.1 Determine the model blending criteria
    • STEP 2.1.1 obtain LAM maximum reflectivity:
    • STEP 2.1.2 obtain Welch model power over latitudes
    • STEP 2.1.3 Obtain the large scale power ratio over the total power:
    • STEP 2.1.4 Obtain the wet ratio over lands
    • STEP 2.1.5 Blending criteria
  • STEP 2.2 Load model data, the model errors and auxiliary files
    • STEP 2.2.1 Load model forecasts
    • STEP 2.2.2 Compare the current and historical power differences between LAM and global models
    • STEP 2.2.3 Load model errors
    • STEP 2.2.4 Load user defined error ratio profile and update model errors
    • STEP 2.2.5 convert model errors from an array to a matrix
    • STEP 2.2.6 create the final error matrix:
  • STEP 2.3 Best Linear Unbiased Estimator
    • STEP 2.3.1 Obtain the model error coefficient
    • STEP 2.3.2 Smooth the model error coefficient over levels:
  • STEP 2.4 Running analysis
    • STEP 2.4.1 obtain the current FFT of LAM and global forecasts level by level:
    • STEP 2.4.2 running analysis:
    • STEP 2.4.3 Difference between WRF, GFS and the blended analysis