Numerical Weather Prediction

Numerical Weather Prediction

We provide expertise for the integration of NWP models in full operational chains.

request a quote
Weather

Numerical Weather Prediction
(NWP) Modeling

A consequence of global warming determined by climate changes is a significant increase in the number of extreme weather events and their intensity, with a huge economic impact and loss of lives: tornadoes, severe thunderstorms, flash floods, wild-fires, winter storms, drought, extreme precipitation, hail. Most of these phenomena, due to their horizontal scale, are considered mesoscale (2-2000Km) and microscale (< 2km) phenomena. Most of the national meteorological agencies are interested in increasing their capability to predict such phenomena and developing efficient early warning systems.

Early warning means Early action.

Based on our more than 25 years of experience in NWP modeling and integration of NWP models in full operational chains, we are now offering the following services:

  • Installing and configuring an NWP model on a dedicated cluster (preferably WRF)
  • Integrate the NWP into the operational chain
  • Including locally available data in the Data Assimilation system of NWP (preferably WRFDA)
  • Configure the Post Processing system
  • Implement a system for output data verification
  • Professional consulting for the fine-tuning process of NWP in order to achieve the expected results (depending on the climatic complexity of the region, this process could take longer than one year)
  • Implement an early warning system based on the NWP nowcasting and sort range forecast output (radar information being already included during the data assimilation process)
  • Implement the End User Lab - a GoForMet module dedicated to end-user product creation.

From a series of available NWP models, we propose our customers use WRF from NCAR, being widely used and well documented, and we will detail our services taking the WRF as a model, but we do not limit our offer to it.

Implementing an NWP model

One of the most well-known and widely used weather forecasting models in the world is the WRF (Weather Research and Forecasting Model), a state-of-the-art atmospheric modeling system. Developed by NCAR, WRF is in the public domain, and it is freely available for download and use (code and documentation) by any public or private entity that has the necessary financial means and professional skills. Suitable to be used for many applications at different scales, from Microscale and Mesoscale Meteorology to global-scale modeling, the WRF package also includes the WRF Data Assimilation System (WRFDA), allowing users to improve the nowcasting and short-range forecast with the data available locally.

WRF is mainly used as a regional Numerical Weather Prediction (NWP) model to improve the forecast provided by the global models with a coarse resolution by running at high resolution on one are several nested domains. Practically, it acts as a dynamic adaptation of the worldwide weather prediction to the specificity of the local micro-climate. Even without improving the initial conditions by assimilating local data, only running the model at higher resolutions allows it to take into account better-defined surface fields (e.g., elevation and waters) and to include in the run parametrization of mesoscale and microscale weather phenomena as tornadoes, thunderstorms, extreme precipitations, etc.

After installing the software on a cluster (a server with several nodes) and configuring the domain of interest, a fine-tuning of the NWP model is absolutely necessary in order to improve the forecast of the original global model. For this activity, experienced professionals with a good understanding of the physics of the model are requested. Depending on the complexity of the terrain and the variety of climatic zones in the selected domain, accomplishing this goal could take from several months to even more than one year.

Only after reaching this first goal can we move on to the next step, namely improving the very short-term forecast by enriching the initial state with the data collected locally (AWOS, radar, satellite, precipitations).

The WRF run (if not used for global forecast) is mainly driven by the initial and lateral boundary files provided by the output of a worldwide model run (GFS), and even including the data assimilation in the operational production chain, only the very short range forecast will be improved, but this is still very important because it is the main tool for nowcasting, and the efficiency of an early warning system is dependent of the quality of nowcasting and short-range forecast.

The most difficult part of data assimilation implementation is to develop modules able to convert all the locally existent data in a variety of formats in the standard one accepted by the WRFDA module and to integrate these modules into the operational chain.

Specialized model data visualization and objective verification tools could be very useful during the entire process of implementing an operational forecasting chain based on WRF, and the quality of these tools could speed up the process.

Use cases

  • Global WRF runs, but this is not a commonly-used configuration in the model. It is preferable to use data for another global model run (GFS)
  • Running operational a regional WRF on nested mesoscale domain
  • Running WRF on one or more nested microscale domain(s) for research purposes

Implementation of an operational chain

The implementation of an operational weather forecasting chain based on the WRF model includes the following steps:

  • Acquisition of the hardware: a cluster with a server and several nodes - dimensioned and configured according to the operational needs based on WRF system requirements
  • Install all the necessary libraries and the WRF itself
  • Configure the NAMELISTs parameters

Run the WRF based on fine-grid initial and boundary condition files, using 3-D meteorological fields interpolated from the coarse domain and static, masked, and time-varying surface fields from the higher-resolution static fields (such as terrain, land use, waters, etc.)

  • In the absence of a locally global run of WRF, the fine-grid initial and boundary condition files need to be downloaded from another center (GFS). Usually, in this case, the output from the global run is available just 6h later than T0 time, which means that the fine-grid initial file for the mesoscale WRF will be the T0 + 6h of the global model.
  • Verification of output from the mesoscale WRF model against the output of the global model

Fine-tuning of NAMELISTs parameters is mandatory to get improved results.

  • This is the most difficult part of the implementation process and depends on the complexity of the orography on the chosen domain, the dimensions of the domain, and the variety of the climatological zones inside the domain. The last three steps (configuring parameters, run, and verification) are repeated until obtaining satisfactory results.
  • Only after accomplishing the previous steps can one go further to integrate the data assimilation module WRFDA in order to improve the fine-grid initial file with data available locally.
  • Installing all the WRFDA libraries for 3DVAR

Convert the observation data from various formats (ASCII, BUFR, MADIS, HDF, etc.) in LITTLE_R format, accepted by the observation preprocessor module.

  • The radiance data can be ingested by configuring inputs for different satellite instruments in several formats (BUFR, HDF, HDF5)
  • Doppler radar data velocity and reflectivity could be assimilated after prior data preparation and appropriate WRFDA configuration.
  • Configure the NAMELISTs parameters to get the desired configuration, and prepare the initial files for WRFDA: first, the guess file, observations, and the background error statistics file.
  • Running the WRFDA will combine a WRF file with observations and error information to produce a “best guess” of the atmospheric state for the WRF operational run. All the WRF lateral boundary conditions will be consistently updated as well.

Ensemble simulations are used to represent the model uncertainty by applying a small perturbation at every time step to each member and can be optionally used in the operational chain.

  • If an ensemble of forecasts is available, the WRFDA can also include the Ensemble Transform Kalman Filter (EKTF) as a hybrid ensemble-variational (EnVar) data assimilation technique, ETKF system updating the ensemble perturbations.
  • We recommend implementing this advanced configuration only after gaining enough experience with the operational chain using the common WRF 3DVAR schema.

Our experience

Our team has professionals who: