Weather in the Tropics
Large forecast producing actors (such as NCEP, ECMWF and UKMO) have for a long time provided high quality forecasts to their general public in the US and in Europe, which both are located at the mid-latitudes. Naturally, since none of the actors has as its core mission to make tropical weather forecasts, this has been of secondary rather than primary interest to develop. The underlying dynamics causing most of the day-to- day weather at these mid-latitudes is represented by quasi-geostrophic turbulence theory (Charney, 1971). This theory (and the relevant equations, i.e. the primitive equations set, that are numerically solved) accurately predicts the large scale evolution of the weather, e.g. in predicting low- and high pressure systems, frontal systems, airmass advection and associated parameters such as temperature, humidity and frontal rain distributions. Even as far out as at 5-7 days, forecasts are of generally good quality thanks to the predictability of the synoptic and planetary scales. And in fact, as much of the weather experienced has its origin in these large scales, forecasts are of satisfactory quality 5 days out most of the time.
The problem arises when there is a public perception that the global models are equally skillful in predicting weather in the Tropics. Many weather forecast providers’ even present forecasts as far out as 14 days or more through web and mobile applications with apparent detail. However, such provision is not backed up and cannot be defended from a scientific aspect. The reason is that most of the day-to-day weather in the Tropics has its origin in the small-scales where convection is a driving force. The spatial and temporal distribution of convection is inherently difficult to predict due to its stochastic nature, while the envelope of convection (larger scale convective patches) can be predicted fairly well to some degree, much like the synoptic scales at the mid-latitudes. In the Tropics, the quasi-geostrophic approximation collapses at scales smaller than planetary due to the proximity to the equator such that ageostrophic effects become of greater importance, while driving mechanisms at the mid-latitudes such as temperature and pressure differences are not present or of orders of magnitude smaller. Therefore, there is no a priori reason why the global models should perform well in near equator areas. Rather, without any compensating algorithms to predict tropical weather with detail, persistence and statistics-based forecasts would likely perform as good or even better.
Moreover, much of the initialization of a forecast model run, i.e. collection and processing of measurement data, are heavily relying on ground based weather observations from synoptic weather stations. Such infrastructure is severely lacking throughout the Tropics, and given the small-scale nature of many weather events, important data is never retrieved such that large errors are present already at the start of a numerical forecast model run which will quickly degrade the forecast quality. In recent years, satellite retrievals are becoming a much more viable choice for initialization, also thanks to the relatively higher resolution of geostationary satellite data sets in near-equator areas.