How it's made: The Short & Sweet Guide to Statistical Post Processing

Welcome to the third part in the How It’s Made series -  Category Three: Statistical Post Processing.

How It’s Made is a 20-part series, going under the hood of weather forecasting. Through the series, we are exploring the Five Categories that create an accurate, reliable forecast.  This post is an introduction to Statistical Post Processing, revealing what it is and how it helps to make your weather forecast.

Download your copy of "How It’s Made: The Ultimate Guide to Weather Forecasting" below:


Download Now

 

Statistical post processing is a technique used by weather experts to enhance and improve their forecasts. It’s an umbrella term, describing multiple statistical methodologies, which each have a unique purpose and application.

 

It works by combining metocean model data and high-resolution environmental data, exploiting the strengths of each. It also applies locally-observed weather from observation networks. Post-processing is used to correct the quite coarse-scale nature of model output; these corrections are necessary to ensure that local effects are taken into consideration and the most accurate forecast is produced.
 

"For site-specific forecasts, raw model data is not accurate enough. By applying smart statistical corrections, based on local observations and terrain data, our forecasts gain about 20% in accuracy"

- Wim van den Berg
Senior Meteorological Consultant
WeatherTech Team
MeteoGroup

 

The four main flavors of statistical post processing


Weather experts produce their core forecasting systems by taking a mix of data from relevant weather models and performing statistical post processing. There are four main flavors used, all aimed at different use cases.

 

#1 Model Output Statistics (MOS) & MeteoBase

MOS - also known as the MultiModel-MOS as the technique is applied to several weather models -  is the weather experts’ showpiece. It’s used to correct for local influences and, as a result, get closer to the actual situation. The MOS takes account of two years of historical data and compares local observations with the model data issued. MOS techniques add value to the raw model forecast, especially for “local” weather parameters like temperature, wind speed and amount of rain.

Each location has its own precise local characteristics which will not be fully reflected in a coarse-scale model. The MOS realigns the deviation and ensures that the appropriate weighting and characteristics are applied from the three main models. As such, a specific individual MOS forecast is generated for each MOS observation point.

 

How it is used

MOS provides data for data feeds and APIs, which are used for a wide range of applications, including crop forecasting in the agriculture sector and reporting on extreme weather for insurance companies. It provides input for more specialist systems, like Road & Route models. It also underpins specific weather products, such as Dynamic Line Rating.

 

How the experts add value

By using two years of training data, experts ensure forecasts are adjusted to local observation sites. As well, by combining data from several models improves skill. Experienced meteorologists put the finishing touch to the MOS forecast, allowing for extreme weather conditions. This adjusted version is called the MeteoBase.

 

#2 Nautical MeteoBase (NMB)

NMB is a forecast data engine that feeds all marine-related products (both shipping and offshore). NMB draws on various sources of atmospheric and oceanographic model forecast data.

For critical marine operations, the weather experts also provide derived elements like risk wind speed and risk wave height. The output data is refreshed 4 times each day.

 

How it is used

NMB is specifically for marine applications in Shipping and Offshore. It feeds into specialist products like SPOS and NowcastingPro. It also underpins custom route advise for master mariners and custom Metocean reports.

 

How the experts add value

Weather experts improve the forecast by combining data from different weather models, including ECMWF, UKMO, KNMI, and NCEP/NOAA. Each model is given its own weight, based on its relative performance. The weights are variable depending on the forecast lead time.

Through model mixing, NMB enhances the strong points of the input models and reduces the weak sides, which improves the accuracy and reliability of the forecast.

 

#3 Scalable Downsizing (ScaDo)

The weather model data used for forecasting can cover large grid areas of up to 50km. As a result, it doesn’t always consider the impact of features that fall within those grid lines, such as changes in elevation and land-use. Downscaling helps to overcome this problem. During the process of downscaling, data from the large grid areas is enhanced with geo-statistical methods to make predictions about weather and climate ‘on-the-spot’, particularly at the surface level.

The ScaDo forecasting system combines the quality of station-based MOS with information about land topography and use, which makes it particularly effective in locations without any observation stations nearby.

 

How it is used

ScaDo is invaluable for locations that experience highly localized weather effects, such as an Alpine valley. It provides detailed wind forecasts for wind farms, feeds into the route based forecast model and can run along a predefined list of points like a powerline.

 

How the experts add value

ScaDo provides forecasts for locations with no observation and improves forecasts in valleys and mountains. It offers customer more data points and means that customers working in complex terrains can still access accurate weather forecasts.

 

#4 Road and Route Models

The road surface model combines a physical and statistical model, designed to calculate the forecast for road surface temperature and conditions at specific locations, such as Road weather information system (RWIS) locations and downscaled locations (i.e. ‘on-the-spot’ forecasts). The route based forecast model is a physical model, designed to calculate forecasts for road surface temperatures and conditions for predefined gritting networks (or routes).

Both models draw on data from MeteoBase, which contains manually controlled output from the MOS forecast system.

 

How it is used

The Road & Route models are used to identify situations where the weather poses a risk on the roads, supporting gritting decisions. The automotive industry also uses them as inputs for autonomous driving initiatives.

 

How the experts add value

For the road surface model, the experts combine the physical model data with 3 years of historical statistical data, which takes into account the local environment of an RWIS station. This improves the model’s accuracy. They also run a bias filter that looks at the past 21 days to identify any inconsistencies.

For the route forecast model, they apply scaling to the physical model to improve the spatial resolution; for example, by adding information about the temperature and condition of road surfaces between the RWIS sites.

 

What are the benefits of statistical post processing?

 

Statistical post processing is an essential component to create your forecast. It enables the weather experts to refine the quite coarse scale natural of model outputs and correct for local variations. This improves the accuracy of the forecasts and helps customers working in their specific operational environments.



Download your copy of "How It’s Made: The Ultimate Guide to Weather Forecasting" below:


Download Now

For site-specific forecasts, raw model data is not accurate enough. By applying smart statistical corrections, based on local observations and terrain data, our forecasts gain about 20% in accuracy


Wim van den Berg
Senior Meteorological Consultant
WeatherTech Team
MeteoGroup