WRF Hindcast

From Agrineer.org Wiki
Revision as of 14:41, 28 June 2017 by Greg (Talk | contribs) (Platform)

Jump to: navigation, search

Agrineer's WRF Hindcast Project generates the input data used by the Grow Degree Calculator (GDC) and the Soil Moisture Estimator (SME) applications. WRF stands for Weather, Research, and Forecasting and is a program made available by UCAR/NCAR and other research participants.

WRF is normally used to forecast weather, but for our purposes we use it to simulate weather in the past.

Sectors

WCONUS sector grid, current operational sectors are shown in green.

The area modeled by the WRF program is called a "domain", but since this implementation uses three nested domains (at 30km, 10km, and 3.3km resolution), the area of interest, the third domain, is called a "sector". Each sector is made up of 171x171 "pixels", 3.3km x 3.3km in size. Client applications can access the pixel's simulated weather data by indicating latitude and longitude coordinates.

The image to the right depicts the sectors defined, and currently active (operational) sectors are shown in green. The defined set of sectors are designated WCONUS (Western CONtinental United States) with an appellation of grid location. Sector rows are given letters and columns are given numbers, with the origin on the bottom right. For example, the bottom right sector is WCONUS_A0 and the top left sector is WCONUS_E4. Grid coordinates are based on data delivery needs and computation capacity. The most eastern sector users will want data earlier than the western ones, by about two hours, and so computation starts with "0" column. Likewise, the most southern sectors will have an earlier planting season than northern ones, and so the most southern row is "A".

Implementation

The WRF climate simulation program requires a sophisticated input stream, consisting of daily data and configuration files, and is usually executed on a parallel computing platform. See Platforms below for Agrineer's implementation. Each sector's output data is calculated off-line and delivered to the web server on a daily basis. Users can expect a maximum of one day delay, with eastern sectors getting results in the morning and western sectors getting results later in the day. Given enough compute capacity it would be possible to deliver results for all sectors in just hours.

Output

Typical WRF output include hundreds of variable values amounting to about 4.5 GB (1.5 GB per domain) per sector daily. A portion of these are used to calculate the standard reference evaporation (ETo) values in the SME application. Since the daily ETo are calculated off-line using the WRF output, we could deliver to the server just the ETo and precipitation values for accumulation purposes. However, the GDC application requires daily maximum and minimun temperatures and so we deliver these as well in netCDF format. All three domains are available on a daily basis from here.

Global Forecast System

Agrineer's implementation of the WRF program uses the Global Forecast System (GFS) forcing files as input. These are files generated by NOAA's National Centers for Environmental Information (NCEI),which are used to force data interpolations to defined values at certain times, projected into the future at six hour intervals. Once a projected time is reached, the forcing file is regenerated using real data instead of projected data and a new set of projected files are generated, and the process repeats itself. This project uses the re-analyzed files at six-hour intervals for historical evaluation. The resolution of the data is 2.5 x 2.5 degrees, and the gridded data is interpolated down to a 0.25 X 0.25 degree resolution to provide finer detail. The data is retrieved from here where the folders are date and interval dependent. To retrieve re-analyzed data for 2017/03/03 interval 00, for example, go to gfs.2017030300, then find gfs.t00z.pgrb2.0p25.f000.

Platform

Agrineer's WRF platform is Linux Mint as the operating system on an 8-core 64-bit AMD based computer, compiled under GNU gfortran with MPICH parallel implementation. The Python language is used to integrate all of the working parts, from retrieving input files to uploading results to the server. The dynamic solver model used in WRF is the ARW (Advanced Research WRF) core.

The computer platform described can calculate 10 sectors per day, with rest periods for maintenance. For redundancy, Agrineer operates two off-line WRF platforms in separate long distance locations.