We all know the devastating power of hurricanes, typhoons, and their Southern Hemisphere counterparts. It is crucial that we predict their behaviour accurately to avoid loss of life and to better guide large-scale infrastructure operations. Although tremendous progress has been made, especially in predicting their propagation path, the intensity or wind forecasts are much more difficult. This is related to the fact that the path of a hurricane is largely determined by the large scale atmospheric environment, and we know that environment quite well. However, intensity has to do with small-scale details in the core regions of hurricanes, and these are much harder to predict. The largest unknown is the mysterious rapid intensification, in which the wind speed in a hurricane can increase from 50 km/h to an astonishing 300 km/h in two days.
Figure 1: a) Satellite view of Hurricane Patricia just before landfall, and b) maximum wind at 10 m above the sea surface in Hurricane Patricia (Note 1 m/s corresponds to 3.6 km/h).
Hurricane Patricia (see figures 1a and b) in 2015 holds the rapid-intensification record and we have studied her in detail. Fortunately, we had an exceptionally detailed data set of temperature, humidity and wind fields in the inner region of the Hurricane from aircraft measurements. (Indeed, they did fly the plane straight through the core of the Hurricane…) This provided an unprecedented view of the inner structure of the Hurricane, but also allows us to study the influence of these observations on prediction.
For this prediction we update the model fields, such as the temperature field and the wind field, using a technique called data assimilation. Data assimilation is a systematic method to incorporate observations into computer models (see e.g. the open access book Evensen et al, 2022, with over 40,000 downloads). For the results below we use a state-of-the-art Local Ensemble Transform Ensemble Kalman Filter, abbreviated to LETKF (see Tao et al. 2022 for details of this study). We run two experiments, one in which we assimilated only large-scale satellite data, and one in which we added the aircraft data of the inner hurricane regions. This resulted in two forecast ensembles, the yellow-brown lines and the blue lines in figure 2.
Figure 2: The strength of the wind as function of distance to the centre of the Hurricane. Data from two forecast ensembles, one ensemble based on only satellite data (yellow-brown) and one ensemble based on both the satellite and the aircraft data (blue). The purple lines are not important here. Note that the aircraft data give rise to much higher velocities because they resolve much smaller scales.
Figure 2 shows that the ensemble based on the aircraft data (blue lines) shows much higher wind speeds, and these hurricanes all develop a rapid intensification phase and become major category 5 hurricanes. The yellow-brown lines do not use the aircraft data, have much lower wind speeds, and do not develop into strong hurricanes. We conclude that the detailed data in the inner part of the Hurricane are crucial for a proper prediction of the intensity of Hurricanes.
These model predictions can be studied further using techniques from causal discovery developed for Hurricane dynamics (Van Leeuwen et al. 2021). Causal discovery methods try to find cause and effect relations in hurricane evolution. The weaker Hurricanes that do not develop rapid intensification have different connections between the temperature and the wind fields than those hurricanes that do show rapid intensification. Specifically, what is needed for rapid intensification is a collaborative action of the temperature and humidity at the sea surface, strong upward motion in the core region, and rain and snow formation in the region close to the centre of the Hurricane, as well as strong heating of the centre region from the stratosphere. All these work together to heat up the core region of the Hurricane, which provides the energy to increase the winds. These winds bring in more humidity near the sea surface, leading to more rain and snow formation, leading to further heating etc. If all these processes work in Harmony rapid intensification is the result. In contrast, when one of these processes is out of sink, as with the yellow-brown lines, the Hurricane does not grow fast and rapid intensification does not occur.
Concluding, although our understanding keeps increasing there are still many missing parts. One way forward is to use better ways to bring the observations into the prediction models. The methods used today, such as the LETKF mentioned above, are based on linearizations that do not allow us to extract all relevant information from the data. This can lead to incorrect interpretation of the causal relations between hurricane variables. New fully nonlinear data-assimilation methods have been developed (e.g. Hu and Van Leeuwen, 2021) and we are working on implementing these in Hurricane prediction models to improve predictions and to understand these major ‘freaks of nature’ better.