RAL has been a leader in the development of intelligent weather prediction systems that blend data from numerical weather prediction models, statistical datasets, real-time observations, and human intelligence to optimize forecasts at user-defined locations.

The goal of these systems is to reduce the inherent forecast error associated with Numerical Weather Prediction (NWP) models and simplify the forecasting process for decision makers. By using machine learning to understand the error characteristics of the models, we can combine them together to create an optimized consensus forecast. The Dynamic Integrated Forecast System (DICast^{®}) was developed at RAL and is an example of this technology. The DICast^{®} system combines NWP model data and observations to produce tuned forecasts of sensible weather elements as well as other derived or customized weather elements. The system is completely automated, updates as frequently as necessary and produces forecasts out to customized forecast extents and temporal resolutions. The DICast^{®} system has been used as the foundation for numerous RAL projects including the areas of lay forecasting, transportation, wind energy, solar energy and agriculture.

While the DICast^{®} system produces deterministic forecasts, another focus of this project area is to produce some sense of the uncertainty of the forecast, that is, to produce probabilistic forecasts. There are many ways to create probabilistic forecasts from ensembles, but that can be expensive and time-consuming. Ensembles don't always produce the spread-skill relationship that is desired either. One approach that bypasses these problems is to use a technique called Bayesian Model Averaging (BMA).

With BMA we can use a fairly short (30 day) history of freely-available model data to produce an expression of the forecast uncertainty by creating Probability Density Functions for each model component. We can then apply these PDF's on future model runs to produce probabilistic forecasts. The BMA system developed at RAL creates probabilistic forecasts of temperature (and other) variables at arbitrary observing locations and forecast extents with excellent spread-skill relationships. The system is completely automated and produces updates as frequently as necessary.