Here I was, thinking that the data the Norwegian meteorological institute (NMI) uses to forecast the weather for the 70,000 places it brags about forecasting via the website yr.no was collected by charming geezers in wayward places, who empty little cups marked in tenths of millimeters to say how much rain has fallen and observe the wind sock next to the chicken coop to say which way and how hard the wind is blowing. But it’s not cups and wind socks. It’s more like fishing nets.
I got curious about the weather, that is, the weather forecasting, after reading in today’s local paper that areas that rely on cabin rentals as a source of income are complaining about yr.no’s forecasts. The complaint is that the forecasts are so pessimistic that tourists cancel their cabin stays. Rain or cool weather is often forecasted by yr.no but the reality is sunshine or milder temperatures. Yr.no had an explanation: They don’t actually measure the weather in all areas on their list of 70,000, but make a sort of average based on the grid points surrounding the area. The grid points are set up by the numerical weather prediction model HIRLAM (High Resolution Limited Area Model). HIRLAM is based on a consortium of the meteorological bureaus of nine European countries, including Norway’s. Here is a sample of various data at the grid points for Sweden.
The Norwegian meteorological institute explains HIRLAM like this (my translation based on their page in Norwegian): Imagine casting a fishing net where the size of a mesh hole is 10 square kilometers. Where the threads of adjacent holes meet is a grid point. At each grid point, barometric pressure, humidity, wind and temperature are measured over time. There are layers and layers of these “fishnets”, going up as high as 50 km in the atmosphere, and there are two mesh sizes: approximately 10 km (0.1 degree of horizontal separation) and approximately 50 km (0.5 degree of horizontal separation). Met.no says:
Because the atmosphere follows the laws of physics, we can calculate how changes in each grid point will influence surrounding grid points. It is an enormous math problem. To solve it, we need a computer – a super-computer as powerful as possible.
Predictions can be a bit off for weather systems that are less than 10 km wide (like a local thunderstorm), or for areas on the map that fall between grid points. The weather bureau interpolates the data in such cases.
But yr.no had another problem, which wasn’t with the accuracy of its predictions, but how they were presented. yr.no has now realized that people ignored the text forecast, and focused on the symbols. Showing a rain cloud in the graphics for a minimum of rain was throwing (and putting) people off. Yr.no changed an algorithm in their computer program so minimal rain will now still show a sun or cloud with sun. That’ll bring back the cabin renters.
In spite of ever-increasing computer power and ever-increasing amounts of data, weather can still not be predicted beyond ten days ahead and no prediction beyond five days can be considered reliable. The reason for that is the inherent chaos of atmospheric behavior. That butterfly effect, you know. Fishnets can’t catch butterflies.