A GRIB file is a file based on the WMO GRIB standard for encoding gridded fields. There are two versions of GRIB files, version 1 and 2 - LuckGrib supports both. GRIB files typically contain weather information and other environmental information such as wave state.
There are a number of Environmental Modeling Centers (EMC's) which generate weather information, such as the U.S. National Center for Environmental Prediction (NCEP), the Canadian Meteorological Center (CMC), the UKMET service, the European ECMWF among others. NCEP and CMC have made their products freely available.
The GRIB data generated by one of the ECM's typically contains information on many different parameters (for example: wind; pressure; rain; and temperature.) Each parameter may be present at different levels of the atmosphere, for example, consider wind at the surface through to the upper atmosphere. Additionally, each set of parameters and levels will also be available at different time steps - GFS produces forecast data for every 3 hours through to 16 days. The number of different parameters, levels and time steps available will vary from model to model.
NCEP creates a number of operational and experimental products - the main product produced by NCEP is the GFS model. GFS seems to have the widest variety of parameters, levels and time steps of the publicly available models.
In very general terms, the different ECM's each have their own computer simulation of the environment along with a suite of algorithms based on their own interpretation of the physics involved in the changes to an atmosphere over time. The simulations are initialized with a set of initial conditions, which are the most recent set of observations of the atmosphere available. Given these initial conditions, the ECM's computer simulation is run which generates the set of parameters at the various levels at each time step.
The output of these models will have no human review or correction. The GRIB data received by the public is the direct output of the model. By contrast, the official forecasts generated by an organization such as the U.S. National Weather Service (NWS) will take into consideration all of the available weather models along with a variety of other sources, such as satellite images. The professionals then apply their considerable experience to study all the various, possibly conflicting, pieces of information to generate their own forecasts. A single GRIB file is useful, but you need to be aware that the model is just one estimation of what may happen in the future.
As you would expect, the area of numerical weather prediction is an active research area with many people studying and improving each of the different models. There are studies which compare the skill levels of the different computer models - while some of the models do have higher general skill levels than others, none of the models are able to perfectly predict the weather.
Generally, the weather forecasts will be fairly accurate over the short term, perhaps one or two days. As the forecast progress further in the future, the accuracy of the prediction may decrease quickly. Many people feel that forecasts beyond four or five days have little predictive value, although how accurate a GRIB forecast is depends greatly on the type of system being forecast. Storms, fronts and hurricanes are examples of systems that may be poorly forecast in a GRIB file. Where there are alternative source of information available for these features, you should access them, along with the GRIB data.
There are also effects related to land which can introduce uncertainty. For example, the wind forecasts along shorelines may be inaccurate due to their being influenced by local effects (gap effects, wind shadows, etc.)
While you are able to download GRIB files with forecasts out to 16 days, the portions of the forecast which are beyond 7 to 10 days may only be useful in studying the model itself - a 10 day or longer forecast has little value in predicting the actual conditions to expect at those time intervals.
A good exercise when starting to work with GRIB files, or when working with a new GRIB model, or examining weather in a new area, is to download a 10 day forecast and then update it every day. After a few weeks, have LuckGrib show the most distant estimate of the original forecast and then step through the newer files until you are looking at a 1 day or newer file. (Simply use the up/down arrow keys to step through the files.) You will notice that all of the different forecasts for the same day and time may vary quite drastically. This is a valuable lesson to learn - don't put too much trust in long term forecasts.
Some of the models available are based on an ensemble approach - LuckGrib supports one ensemble model, the CMC model from Canada. In a traditional approach, the initial conditions of the atmosphere is gathered, the computer forecasting software is run and a single set of data is created to describe the forecast.
An ensemble model alters this approach slightly. An ensemble model will run the computer simulation many different times, each time with slightly different initial conditions and possibly different model physics. The idea is that if you alter the initial condition, or model, slightly, before running the simulation, then if the outputs of all of the runs are fairly similar, you can have more confidence in the result. If small changes in the initial conditions result in wildly different outcomes, then the system is unstable and it is difficult to know what the true forecast should be.
There is an interesting discussion of the ensemble approach here.
One ensemble model, the CMC model generated by the Canadian government, runs the computer simulation 20 times. Professional forecasters are able to download the result of all 20 computer runs and compare them. The CMC model also generates a set of averages. When you hear someone referring to the "CMC ensemble" or simply the "CMC" data, they are likely referring to the average data created by the ensemble. Some people prefer to work with ensemble average data, believing that it is more accurate than a single non-ensemble run.
And with thanks to...
All of the GRIB models in LuckGrib are being downloaded directly from NCEP/NOAA. Many thanks to that group for making all of their weather information freely available - they are providing a tremendous service to the community of people interested in weather prediction.