Good Model Gone Bad

So there’s no denying that one of my earlier posts ‘Advanced Warning – get shovels and sledges at the ready’ was slightly off the mark when it comes to the weather in the UK this winter so far. Since October I’ve been ready to get my favourite fluffy hat, scarf and mittens out to wear, but they just haven’t been needed. More like raincoats and wellies, than shovels and sledges.

Published by NOAA in December 2015, the months of January to November of 2015 were globally the warmest on record for such a period in 136 years (see Table 1 below). This record-breaker goes across both land and oceans. The average global land surface temperature for the first 11 months of 2015 was at 1.27°C above average in the 136 year period of record. Similarly, the oceans saw an average global sea surface temperature at 0.72°C above average. Finally, the Earth’s land & oceans combined temperature was at 0.87°C above the 20th century average of 14.0°C. For the whole of 2015 not to become the warmest year on record since 1879, the December global temperature would have to be at least 0.81°C below average. If temperatures we’ve been seeing in the UK are anything to go by, this is very unlikely to say the least!

Table 1: Global Temperature Anomalies

Source: NOAA National Centers for Environmental Information

These record high temperatures were most prominent in the majority of land in South America, as well as much of the Earth’s oceans, including; the eastern & central Pacific Ocean, a large proportion of the central western Atlantic, and most of the Indian Ocean. However, there were areas of the oceans with record low temperatures, specifically the southern tip of South America, as well as an area in the Atlantic Ocean, both experiencing lower than average sea surface temperatures. These land and ocean temperatures, as previously determined, are undeniably linked the El Niño weather phenomenon. Perhaps unsurprisingly, El Niño has been hitting the headlines in a big way again over the past few weeks.

What with record high temperatures, combined with extreme precipitation, (there was more than 200% of the average rainfall for November in much of south-west Scotland, north-west England and north Wales), the result has been that of Storm Frank. Storm Frank has devastated many areas of the UK by causing them to experience severe flooding. The floods have resulted in thousands of homes to be flooded, people having to be evacuated from buildings, washing away roads and bridges, and causing potentially greater than 5 billion pounds worth of damage.

PRENSA MUNICIPIO/AFP/GETTY IMAGES

Source: The Uruguay River in Concordia, Daily News

Without undermining the severity and importance of the unfortunate situation these thousands of people are going through in the UK, over in South America in excess of 150,000 thousand people have had to flee their homes due to intense flooding. El Nino has caused extreme rainfall resulting in 3 major rivers, the Paraguay, the Uruguay and the Quarai, to swell and break their banks. The 150,000+ thousand people who have been directly affected from the flooding come from Paraguay, Argentina, Uruguay, and Brazil. Four countries which cover a much larger area than the UK, puts the scale of this disaster into perspective.

It seems that predictions of a cold, snowy winter have been met by that of a mild, wet one. So why do models sometimes ‘go wrong’ and make predictions which are far from what the situation turns out to be in reality?

Essentially, because that is exactly what the model results are – predictions. A prediction is something that is likely to occur in the future given data or knowledge of prior outcomes of the system. There is always a degree of uncertainty within model parameters and the forecasts they produce. When it comes to meteorological models, for example, forecasts are made in iterative time steps (as we all know, we generally trust a weather report for today compared to one for two weeks time). So why is this the case? A tiny error in an initial state input, extrapolates up to be a much larger error at a later time point prediction. Modelling is a continuing development. Whether it comes to more recent data being collected and available for use; advances in technology allowing better or different types of data to be recorded; or, even new insights into a system; these result in models being updated or rebuilt to include the additional information.

Source: Science or Not?

The more that is learnt about a complex physical process, the more detail that is available to be included in the model build, thus potentially more chance of errors. These errors can come from either the assumptions made, or even the actual coding of the model build itself. Models must be falsifiable to test if the model produces results in line with that of the real-world situation. However, physically based models derived from scientifically sound physical principles are often found to not be consistent with observations (Beven, 2002). Even if a model is found to be falsified, this can provide areas in which the model needs to be improved, or an insight to into mistakes in previous assumptions made.

Whenever modelling these issues need to be kept in mind, as the mathematician George Box said: ‘essentially, all models are wrong, but some are useful’.