When neighbours become the best weather forecasters

What is it about?

Imagine that you see a dark cloud – rain is coming! – and you notice it moving in the direction of where your best friend is taking a stroll. Wouldn’t you warn him/her beforehand? That’s exactly the aim of this work, but for forecasting clouds. Sensors measuring solar irradiance can be used to detect clouds. Thus, aggregating measurements from a network of sensors in a forecast leads to improvements in its accuracy. If data is recorded every couple of seconds, an individual sensor can better anticipate incoming clouds based on information from its neighbours. the positioning of such neighbours is critical. Since clouds move with the wind, useful neighbours are placed upwind.On the other hand, more distant neighbours help farther forecasts, because the detected cloud will take longer to reach the target. Beyond a certain distance, individual neighbours are not very informative but they nevertheless contribute to better collective forecasts, due to a wisdom-of-crowds effect.

Featured Image

Why is it important?

PV generation is very sensitive to changes in cloud cover and, just from the passage of clouds, a module can vary its output by up to 80-90% in a few seconds. This poses a challenge for the operation of an electrical grid, where generation must meet demand at every given moment. As if this were not enough, these generation ramps are also highly unpredictable. Traditional weather forecasting models have been showing progressive improvements but are far from being able to predict at the detail (at the individual cloud level and the seconds-minutes range). On the other hand, models which make use of information spread out both in time and space deliver very promising results and thus need to be better understood.

Perspectives