On Tuesday the German national team defied several historical precedents in the course of their world cup semi-final with Brazil. Like a tropical cyclone that intersects with an upper atmospheric instability, the conditions aligned for an extreme, and to many, devastating, event. Could this have been expected? Of course in retrospect we have proof that there was the potential for this, because it happened, but that’s not the same thing. There were subtle warning signs: Brazil were missing two of their top players, resulting in [what before Tuesday had seemed to be] an excessively emotional fan response; the German team had been gaining momentum, and their conservative play against France [it is now apparent] suggested they had started to take a long view approach to the tournament; the Brazilian team was playing under massive expectations, and that sort of pressure [although in other cases it has had the opposite impact] can be highly detrimental for a young team. Still, the odds were long that these factors would all combine to produce a record shattering fifteen minutes of play, although that is exactly what happened. If you had gambled heavily on this outcome, you probably wouldn’t have received much sympathy if it hadn’t come to pass. However, if you were a bookie, and had allowed people to place such bets without adequately allowing that the results could actually occur, you would have gone bankrupt and had no one but yourself to blame.
The same challenges are faced when dealing with extreme weather events. I can’t schedule every aspect of my life as if there was going to be massive flood that day, but I want to have a plan if one is possible where I live. It’s not desirable to abstain from planting crops because there could be a drought, but if I own a farm in a region where droughts occur and don’t plan for that possibility, my enterprise is going to be short lived. Predicting these events, however, even days in advance, can be problematic. As with Tuesday’s football match, there can be warning signs, but they typically refer to the potentially of an event, rather than providing a definitive forecast. As well, the ability to identify warning signs is a result of intensive study of past events, and so provides very limited information about unprecedented occurrences. Experts agreed that there was a good chance that Germany could win Tuesday’s match, but who would have predicted the highest scoring semi-final in World Cup history? For models based solely on historical statistics, this would not even be something they were capable of considering. What signs are indicative that something that hasn’t been seen before is going to happen? This is especially problematic for weather events in areas with limited observations, where it is often unclear what would or would not be unprecedented for the region.
The forecasting situation becomes even more complex when we consider it from the perspective of climate. When filling out a prediction bracket for a major sporting event, people typically forecast fewer upsets than they actually expect will happen. The strategy is that since it is hard to guess which upsets will actually occur, sticking to safe bets gives a chance for these events to balance out, and your prediction will be wrong fewer times overall than those of the people you are betting with. Global climate modes work on very similar principles. They include limited simulation of extreme events, focusing on large spatial and temporal scales, in order to capture bulk processes and typical states; i.e., the climate [1]. The question becomes, not “what will the score of this semi-final be”, but rather something like “is there a trend of an increasing number of high scoring matches in World Cup tournaments”. Of interest is what meta-factors within the ‘climate’ of international football would result in more or less of the type of situations we saw on Tuesday. When working with climate models the challenge becomes to pull such information out of climate simulation experiments, even if the phenomena of interest are not directly simulated, both to make full use of the available data, as well as to correctly identify the limitations in its prescriptive ability.
For the Brazilian team, Tuesdays events have been described as disastrous. The transition from being a statistically extreme event to a disaster is a social phenomena. Teams lose matches all the time, and sometimes by larger margins. What makes people consider this outcome to be so catastrophic is the context. This was a team that was at the beginning of the tournament heavily favoured to be the over all winner. Many commentators believe that the prospect of having a winning team was the only thing suppressing the Brazilian populace’s rising dissatisfaction with the expense and disruptions resulting from hosting the World Cup. The social ramifications of this match are what is drawing much of the media attention. This is the same with weather and other natural events. Famines occur when drought conditions occur in agricultural areas, floods, when high water levels reach human habitation and exceed the infrastructure’s ability to divert them. There are earthquakes (and icequakes) in Antarctica, but no one is hurt. A disaster occurs when an individual or group’s well being depended on something that has happened not happening. As such, not only is it important to understand the odds, but also to never bet more than you can afford to lose.
All that aside, quantitative measures such as odds are not the only way to express distributions of potential outcomes. Other description formats include the “narrative approach”, and the current narrative is that on Sunday we will see the world’s arguably best team, against the world’s quantitatively best player. I can’t wait to watch what happens, and the biggest thrill is feeling that anything is possible.
Footnotes:
[1] Although there is an interesting ongoing debate in the community as to if extreme events can be treated as anomalies, or whether individual events can be major drivers on the climate system, and as such their under representation in climate simulations diminishes our confidence in the model representation of mean states.