Can deep learning beat numerical weather prediction?

The recent hype about artificial intelligence has sparked renewed interest in applying the successful deep learning (DL) methods for image recognition, speech recognition, robotics, strategic games and other application areas to the field of meteorology. There is some evidence that better weather forecasts can be produced by introducing big data mining and neural networks into the weather prediction workflow. Here, we discuss the question of whether it is possible to completely replace the current numerical weather models and data assimilation systems with DL approaches. This discussion entails a review of state-of-the-art machine learning concepts and their applicability to weather data with its pertinent statistical properties. We think that it is not inconceivable that numerical weather models may one day become obsolete, but a number of fundamental breakthroughs are needed before this goal comes into reach. This article is part of the theme issue ‘Machine learning for weather and climate modelling’.

Validation of a New Model for the Estimation of Residual Tropospheric Delay Error Under Extreme Weather Conditions

The electromagnetic signals of the Global Navigation Satellite Systems (GNSS) satellites suffer delays while propagating through the troposphere. The tropospheric delay is a significant systematic error of GNSS positioning. For safety-of-life applications of positioning many systematic error effects are either mitigated or eliminated in the positioning solution. Space based augmentation systems provide corrections for the orbital and satellite clock error, the ionospheric effects, etc. Moreover advanced GNSS provide dual frequency code observations for civilian users to eliminate the ionospheric delays caused by the electron content of the upper atmosphere. Nevertheless tropospheric delays are still taken into account using empirical models.For safety-of-life applications besides the accuracy of the positioning, the integrity of the positioning service is an important factor, too. The integrity information includes the maximal positioning error at an extremely rare probability level, called protection level to ensure highly reliable position solution in the aviation. The Radio Technical Commission for Aeronautics Minimum Operational Performance Standard (RTCA MOPS) recommends 0.12 m as the maximum zenith tropospheric error in terms of standard deviation. Previous studies show that this recommendation seems to be too conservative leading to a lower service availability. Therefore a more realistic integrity model has to be derived for the estimation of maximal residual tropospheric delay error.In the recent years many advanced empirical tropospheric delay models have been formulated compared to the one recommended by the RTCA. Recently new integrity models have been derived for estimating the maximum residual tropospheric delay error using numerical weather models under real extreme weather.The aim of this paper is to study the reliability of these models conditions. In order to achieve this, high-resolution numerical weather models were ray-traced using an improved ray-tracing algorithm to evaluate the slant and zenith tropospheric delays with the geographical resolution of 0.1° × 0.1°.