This is discussion notes on WRF installation and configuration for urban and regional scale simulations, based on best practice notes on the WRF forums and our experience while playing with a combination of domain settings and physics settings. These conclusions are based on the best of our case studies and our understanding of how the WRF model works and what is the best combination of options to get it working.
At the end, the best option is for you to run the model as many times as possible to replicate your local conditions.
Last updated: 2023.07.28 by Dr Sarath Guttikunda and Dr Md Rafiuddin
***********
Background links:
- WRF download site (link)
- WRF github site (link)
- WRF installation guide (link)
- WRF forum (link)
- WRF namelist best practice (link)
- WPS namelist best practice (link)
- WRF manual Chapter 5 with physics configuration notes (link)
- GDAS-FNL@0.25deg resolution (UCAR link)
- GDAS-FNL@1.0deg resolution (UCAR link)
- GFS/GDAS of all resolutions @ AWS open data repository (link)
***********
High resolution simulations
The following results and conclusions are based on WRF simulations over India, specifically over 3 urban-centric regions – Delhi to represent the Northern latitudes closer to the high mountains, Hyderabad to represent the plateau and plains of the South, and Mumbai to represent a coastal region. All three present unique challenges with high altitudes and strong monsoons. The model outputs were compared with data from the Indian Meteorological Department (IMD) as day totals for a 1-deg box covering the cities.
Time period for Delhi and Mumbai simulation is July, 2020
Time period for Hyderabad is October, 2020
Delhi 1-deg box is defined as 76.7 to 77.7 longitudes; 28.0 to 29.0 latitudes
Hyderabad 1-deg box is defined as 78.0 to 79.0 longitudes; 16.9 to 17.9 latitudes
Mumbai 1-deg box is defined as 72.5 to 73.5 longitudes; 18.6 to 19.6 latitudes
All the simulations were conducted for 1-month period (31 days) and the WRF model was setup in a 5-day re-initial mode. Meaning, we run the model for 5 days, but save only the last 4 days. Then repeat the same with a start from the day before the end of the last run. This helped with spinning the model and not start with zero or default initial conditions.
The model is nudged with GDAS-FNL 00Z data available at 0.25 degree resolution. For simulations on AWS, data was downloaded from AWS’s open data registry.
All the simulations were conducted in 2 nests – 15km and 3km. Only the 3km (d02) data was used for evaluation. This break-up was a convenient choice for cloud physics (cu_physics), which was applied for d01 and skipped for d02 (which is the grey area for most of the cu-schemes). Since, the initial data is available at 0.25deg resolution, 15km start for d01 is acceptable. According to some Q&A on wrf-forum, even 12km and 9km are also acceptable. Caution that 9km resolution will initiate the cu-grey-area problem.
vertical resolution is set at 41 layers using the following eta’s
1.0000,0.9980,0.9955,0.9930,0.9900,0.9869,0.9833,0.9778,0.9705,0.9603,0.9472,0.9307,0.9110,0.8859,0.8559,0.8212,0.7825,0.7396,0.6939,0.6479,0.6043,0.5629,0.5213,0.4798,0.4381,0.3975,0.3583,0.3192,0.2809,0.2436,0.2080,0.1743,0.1430,0.1148,0.0903,0.0691,0.0508,0.0350,0.0215,0.0098,0.0000 — with the first 8 layers under 500m and the first 12 layers under 1km. Because of the high mountains, it was necessary to increase the number lower layers.
physics combinations that resulted in satisfactory comparisons between the precipitation levels simulated in the WRF model and those obtained from IMD are the following
- cu_physics (=1, KF), bl_pbl_physics (=0, 3DTKE), mp_physics (=16, WDM6)