Forecasting in the past couple of decades has evolved from manually intensive, hand-drawn forecasts to automatically generated, high-resolution, gridded forecasts. During this time, the resolution of numerical weather prediction (NWP) has increased from models routinely run in the 30100-km horizontal resolution range down to 3-12 km. NWP in the 1980s ran out to 48 hours, and now goes out 15 days or more. Ensembles have increased in use and sophistication, and there are plans for 3-km resolution ensembles along with global model ensembles on the order of 13 km.
A recent study at Oxford University estimates that nearly half (47%) of the total US employment (702 occupations) is at risk of automation in the next two decades. They found that the probability for atmospheric and space scientists positions to be computerized is 67%. The accuracy of both the raw model output and the post-processed Model Output Statistics (MOS) has steadily improved and is now arguably the same or better than many human forecasts. This logically leads to the discussion of the pros and cons of forecast automation and the role of the human forecaster in the future.
We believe that the potential for atrophy in the area of science is high. The operational forecaster of the future should understand and harness ensemble theory, statistical analysis, and use of probabilistic diagnostics. Expertise in microscale meteorology and translating that into local impacts will be key. Using coupled models in hydrology, wave generation, and road conditions will become commonplace. Nowcasting will remain critical, so effective use of radar and derived satellite products (even outside of severe convection) should be second nature. A full understanding of local climate will enable us to place forecast events in context. Finally, familiarization with forecast tools and uncertainties at the weather/climate interface is needed.
The views expressed are those of the author(s) and do not necessarily represent those of the National Weather Service.