The unpredictability of tropical cyclones, or, why was my uni closed on a sunny day?
Published 31 January 2014
Tropical cyclone Dylan crossed the coast of North Queensland at approximately 3:30am this morning (Friday, 31st January 2014). An early prediction was for the cyclone to directly hit my home city of Townsville, but in reality the cyclone went far south and Townsville got very little wind or rain. Yesterday, everyone seemed convinced that the cyclone would hit us: even the university campus was officially closed due to the threat of the cyclone. Therefore, it was somewhat surprising to wake up this morning to a lovely calm day with absolutely no sign of inclement weather! In hindsight, the Vice Chancellor’s decision to close the campus seems rather unnecessary, to say the least.
So how did this come to be? I thought that it would be interesting to compare the various predictions with the actual path that the cyclone took. The Bureau of Meteorology’s tracking maps are updated several times a day, but they always replace the old maps on their website with the new ones, presumably to avoid confusion. Luckily for me, the tracking maps were also being shared on Facebook by the Townsville City Council, so I was able to get hold of most of the old forecasts. I downloaded all the old forecast maps that I could find, and overlaid them, one by one, on top of a larger map of Australia. I lined up the maps as best I could and then drew over the top of the predicted paths. This is the result:
Each line represents a forecast trajectory that was published at a particular time. The time of each forecast is labelled at the bottom left of each line. The actual path taken is the thicker red line.
Now, in fairness to the meteorologists who made the prediction, the cyclone’s actual path was within the uncertainty of their forecast. As a scientific modeller, you would be happy with this result because it lies within the error bars. The model worked! Yet, there’s something fundamentally unsatisfying about a prediction that was so imprecise that it triggered a university to close down on a sunny day.
To me, this demonstrates one of the greatest challenges in communicating the results of scientific modelling. The challenge is to distil our results into meaningful messages without losing that sense of caution that results from our understanding of the model. As modellers, we know the intricate details of the model and its assumptions and weaknesses, and we judge its output accordingly. Non-specialists don’t have the benefit of this background, so we have to consider carefully how we communicate. I am not saying that the Bureau of Meteorology failed to do this, in fact, I think they made it very clear that the cyclone’s path could be almost anything. I simply think that this incident should serve as a reminder of the fallibility of predictive modelling. A bit of cautious modesty goes a long way.