Weather Modeling & Research

Limits of Weather Prediction

Forecasting extreme weather—from severe thunderstorms to heavy snow—is one of the most important aspects of numerical weather prediction. Up to this day, however, we do not fully know what forecast problems are potentially tractable, mainly because the limits of weather prediction are not well known. To address the question of what is the ultimate limit of weather prediction, we used a cloud-resolving version of the Model for Prediction Across Scales (MPAS), and investigated the growth of differences between pairs of simulations that are started from slightly differing initial states but are otherwise identical (we call those pairs “twins”).

The evolution of forecast differences between one pair of twins is shown in Figure 1 and can be summarized as follows: initially, the atmospheric state is almost the same, but becomes completely different at the end. Specifically, differences initially grow in narrow areas with thunderstorm activity such as the tropics and midlatitude fronts (Fig. 1a), pointing to the difficulty of forecasting thunderstorms even a day in advance. After a day or two, the differences begin to expand (Fig. 1b--c), and over the following days (Fig. 1d--e), keep growing in extent, magnitude, and scale. In the end, the differences are so stark that the weather pattern in one twin bears no resemblance to the weather in the other twin (Fig. 1f).

Figure 1: Sequence of maps illustrating the growth of differences between a pair of twins over a 20-day period
Figure 1: Sequence of maps illustrating the growth of differences between a pair of twins over a 20-day period. The twins are started from slightly differing initial states but are otherwise identical. Darker colors mean larger differences.


But what is the limit of weather prediction? Figure 2 gives us an idea by comparing the globally averaged difference between a pair of twins (solid line) with a benchmark (dashed line). The predictability limit is reached when the difference becomes equal to the benchmark. This occurs on day 17, indicating that for this particular case, the weather is predictable for a little over two weeks. However, given the limited value of only one case, one should not take the predictability limit of 17 days too literally, and a vaguer statement like "the weather cannot be predicted beyond 2–3 weeks" seems more appropriate.



Figure 2: Time series of globally averaged differences between a pair of twins and a benchmark
Figure 2: Time series of globally averaged differences between a pair of twins (solid line) and a benchmark (dashed line). When the two lines intersect, here at 17 days, the limit of predictability is reached and the forecast is "as wrong as it can possibly get."




Flexibility of the WRF and MPAS-A Modeling Systems

As part of our goal to develop, deliver, and support a suite of advanced atmospheric multiscale community modeling systems, in the spring of 2018 MMM released the capability to drive NCAR’s regional Advanced Research WRF (ARW) forecast model using global forecasts from NCAR’s Model for Prediction Across Scales - Atmosphere (MPAS-A). Depicted in the figure below are 4.5 day MPAS-A forecasts of near-surface water vapor for hurricane Harvey that made landfall on the Gulf coast of Texas in August 2017.

The MPAS forecasts were produced using a global 60-15 km horizontal mesh configuration and a regional MPAS 15-3 km horizontal mesh configuration. A 12-hour forecast using regional WRF employing a 1.33 km grid is also shown in the figure, and this forecast was initialized and driven by the 15-3 km regional MPAS forecast. The relative MPAS-A mesh spacings (white), and locations of regional MPAS and WRF meshes (yellow), are shown as contour lines on the forecast plots. The regional MPAS-A capability is scheduled to be released this coming year. These forecasts highlight the flexibility of the WRF and MPAS-A modeling systems that NCAR develops and supports for the community, and demonstrates that the forecast systems can be used together to simulate atmospheric phenomena from global to convective to Large-Eddy-Simulation (LES) scales.

Figure: MPAS-A forecasts of near-surface water vapor for hurricane Harvey
Figure: MPAS-A forecasts of near-surface water vapor for hurricane Harvey, produced using a global 60-15 km horizontal mesh configuration (left), a regional MPAS 15-3 km horizontal mesh configuration (top right), and a 1.33-km WRF domain driven by the 15-3 km MPAS (lower right).




WRF Community Support

WRF support provides a valuable service to the research and university community. Community publications based on research using the Weather Research Forecasting (WRF) model continue their pace, in part reflecting NCAR’s support and development efforts. WRF, WRF-DA (WRF-Data Assimilation), MPAS (Model for Prediction Across Scales) and RCM (Regional Climate Modeling) tutorials address demands of the international research, applications, and operations communities for learning and applying the models and will contribute to training new users.

MMM hosted the 19th WRF Users’ Workshop in June 2018, with an attendance of over 170. MMM also conducted two WRF tutorials at NCAR (January and July), totaling over 130 participants. Tutorials for RCM and MPAS tutorial also attracted a large number of attendees. The diagrams below illustrate the trend in the number of papers published which used WRF, and the number of WRF registered users.

Figure: MPAS-A forecasts of near-surface water vapor for hurricane Harvey
Figure: MPAS-A forecasts of near-surface water vapor for hurricane Harvey