The Evolution of Forecasting

Humankind has been trying to predict the shifting patterns of the world around us for centuries. Weather forecasting, our way of predicting the conditions of atmosphere for a given location and time, is one of the critical tools employed to do so. As long as we have had these tools, our “environmental psychism” has propelled us forward.

Photo Credit: Stefan Widua

In Ancient times, communities typically relied on observed patterns and recognized sequences of events for weather forecasting. Around 650 BC, the Babylonians, for example, predicted the weather from cloud patterns.

Fast forward to the second half of the 19th century, weather prediction became more based on observational data over wider areas, thanks to the spread of telegraph. In this era of pragmatism, the behavior of the earth and skies were approached only with a limited understanding of the physical mechanisms behind the weather processes. It was only a little later where further advancements revealed the influences of thermodynamics and hydrodynamics on changes in the atmosphere; this lead to the mathematical formulation of atmospheric dynamics in 1904. By 1922, mathematician Lewis Fry Richardson published the first attempt at using numerics to forecast the weather, which was unsuccessful, but demonstrated that in order to compute weather forecast in real time, 64000 people should be solving equations by hand.

The Numerical Weather Prediction (NWP)

NWP was first attempted in the 1920s, but it was not until the advent of computer simulation in the 1950s that numerical weather predictions produced realistic results. NWP is a method of weather forecasting that uses mathematical models of the atmosphere to predict the weather based on present weather conditions. It employs a set of equations that describe the flow of fluids and thermodynamic relations, which then are translated into computer code. Combined with the observational data as initial conditions, the model is then capable of computing the properties of the atmosphere over time, which we usually see as a weather. While this sounds pretty straightforward, actual forecast is almost never perfect and as we all have seen, sometimes just wrong. Without going too far into the details, the three main factors limiting accuracy are:

1. Incomplete observational data
2. Approximation in the equations describing the atmospheric processes
3. Limitations in the computing power.

Since 1950’s, Numerical Weather Prediction (NWP) uses the most powerful computers in the world, to perform the forecast.

The output of a NWP model is a set of weather parameter distributed over some spatial domain, which can be used directly or examined by a specialist to provide a simple, human comprehendable forecasts. Different kinds of NWP models are used, depending on the application and need. There are two major types of models: Global and Regional/ Mesoscale.

The Global NWP Model:

• The domain of the forecast is the whole globe;
• The forecast period is longer, usually 10–15 days
• Forecasts will usually have a lower resolution;
• Utilizes global observation data;
• Examples include: GFS, ECMWF

Global NWP Forecast Model completed with Planet OS

Above you will see a Global NWP Forecast I generated using data that is available on the Planet OS Datahub. In this forecast, we can see the significant heatwave that is affecting Australia right now as well as the cold winter temperatures in the North Eastern United States.

The Regional NWP Model:

• The domain is a specific area (eg. North America, Hawaii, Scandinavia or any country);
• The forecast period is usually 2–3 days
• Forecasts and visualizations will have a higher resolution, which is necessary to predict small scale weather events;
• The domains of the model have edges/boundaries;
• Examples include: NAM models, FMI HIRLAM, Meteo France Arome

Regional NWP Forecast Model completed with Planet OS

Due to the three sources of uncertainty stated above, weather forecasts are not perfect and will not be in the foreseeable future. Because different research groups have distinct approaches to tackle these problems, the models used in various institutions are differ from one another. So it is quite common, that for a particular location, there can be three or more (some global, some regional) NWP model forecast available.

At Planet OS Datahub, we try to make as many different models as possible available, with the same API, so everyone has the ability to utilize and compare the models themselves.

Want to learn more about the Planet OS Datahub? Simply create a free account and try it out! Have additional questions or need help getting started? Contact us via the chat tool on the website or subscribe to our email newsletter to receive future updates in your inbox.