Chapter 8b: Advanced Weather Forecasting

Numerical Weather Prediction (NWP) Model Forecasts

Numerical weather predictions are forecasting techniques that uses computer models to predict the weather. The simulation process begins by entering data about the current conditions of the atmosphere, which the computer then calculates how the conditions will change over time.

Common NWP models operate on grid points ranging from 1 to 80 kilometres apart. Actual weather data is not available for all grid points, so some of the starting data is extrapolated. Finer grids are desirable, but come at the expense of requiring large and expensive computers. Coarser grids are relatively quick and cheap, but may not accurately capture the topography in mountains regions.

Initialization. The process of entering current weather observation data into a computer model is called ‘initialization’. The types of weather observations used for NWPs include:

Atmospheric soundings
Weather satellite data
METAR observations
Reconnaissance aircraft
Ocean surface buoys

The World Meteorology Organization standardizes weather station instruments and observation guidelines, including timing and reporting codes. Data used in NWPs must first be verified and analysed to ensure their quality and quantity to create a regular grid. Very small errors in the initial input data, compounded by areas that have very sparse actual observation points (the Pacific Ocean for example), will amplify with each run of the computer model.

Parameterization. NWP models do not handle small-scale complex processes very well. In order for models to account for these finer details, the processes must be added to the models.

For example, a convective cell may be at a scale of less than one kilometre. A typical model grid box is in the range of 5-300km, therefore in order to account for this small-scale phenomenon, each grid box is assigned a number representing the degree of convective activity. The number is determined from a dedicated routine in the model. In this way, the process on convective cloud formation has been parameterized.

Domain. The domain of an NWP model refers to the horizontal size of the model. A global-scale model covers the entire Earth; a regional-scale model includes a portion of the Earth; and a meso-scale model usually has a fine-scale grid.

Vertical Grid. Some NWP models use a vertical coordinate system that scales atmospheric pressures with respect to the pressure at the surface. Other NWP models use a terrain-following coordinate system (sigma) which conforms to the natural terrain. Sigma models are advantageous in complex terrain because they allow for higher resolution just above the ground level.

Model Output Statistics (MOS). Model output statistics refers to statistical techniques for post-processing output from NWP models. MOS is used both for correcting model forecast variables, as well as predicting variables not explicitly forecasted by the model. In this way, deterministic NWP models are combined with statistical MOS elements to produce a hybrid model containing both deterministic and statistical elements. The accuracy of such a model is generally far better than either a pure deterministic or pure statistical mode.

Metograms / XT Diagrams. A metograms (often referred to as an XT diagram) is a graphical depiction of a collection of weather patterns from an NWP model forecast. It is displayed as a time series, with a horizontal time scale and vertical parameter scale. Metograms typically include:

Tropospheric height cross section with relative humidity
Temperature and wind
Freezing level
Wind speed and direction at specific elevations
Temperature and dew point at specific elevations
Cloud cover and precipitation amount

An example of a metogram produced for Mt. Saint Annie, in the North Columbia Mountains. The metograms shows a tropospheric time-height section; wind speed and direction; cloud cover; temperature and dew point; and precipitation.

Long-Range Weather Forecasts

Many weather forecasting centres provide long-range weather forecasts. Avalanche Canada issues a long-range weather forecast for 3-10 days once per week during the winter season. It is based on a combination of deterministic weather models, model ensembles, and a briefing from a professional meteorologist. It is not intended to accurately predict conditions at any specific time or place. The purpose is to highlight the most likely scenarios for the periods indicated.

Some weather centres also provide long-range weather forecasts based on statistical and climatological data, however these products are at best a guide to the expected average conditions ahead. Daily weather patterns will almost certainly deviate widely from the given output. To the inexperienced, they may do more misfortune than provide useful guidance.

Ensemble Forecasts

An ensemble forecast relies on a collection of outputs from a number of different models. Some ensemble models, like the North American Ensemble Forecasting System (NAEFS), may improve the reliability of long-range forecasts for the 5-10 day period.

There are a number of ways ensemble forecasts can be displayed.

By changing a single parameter value in the ensemble, the mean value of the ensemble can be displayed.
By calculating all possible values for a single parameter in the ensemble, all of the values can be displayed. In this case, the spread of the data gives an indication of the variance, with a high spread indicating a high variance and therefore a low confidence in the forecast.
By displaying the data range on an EPSgram (showing the range of predicted precipitation values, for example), the median value, the interquartile range, and the full data ranges can be displayed.

Generally, when the ensemble of models agrees, the likelihood of that scenario coming to pass increases. With practice, you may be able to deduce other potential scenarios for the same time period. As an example, consider a 10-day forecast based on an analysis of the location of the 546dam/500hPa contour line on a spaghetti plot. Most of the models indicate a blocking weather pattern, which would keep the province dry and cold, but some may indicate a return to zonal flow, which would bring frontal systems and precipitation.

An example of a temperature anomaly chart with explanation notes.

An example of a 546dam spaghetti plot. The image on the left is the 0hr forecast, and the image on the right is the 72hr forecast.

An example of an EPSgram with explanation notes.