search

openclimatefix / power_perceiver

Machine learning experiments using the Perceiver IO model to forecast the electricity system (starting with solar)
MIT License
7 stars 1 forks source link

[Data exploration] Compare clouds in NWPs to satellite imagery #19

Open JackKelly opened 2 years ago

JackKelly commented 2 years ago

Another thing I should've looked into years ago!

Compare the "{low, medium, high} cloud cover" fields in Met Office's UKV NWP to actual satellite imagery.

The video I made last month plotted "downward shortwave radiative flux", not cloud cover. I could modify that same script to animate cloud cover (maybe visualise low, medium and high cloud cover in different colours?)

JackKelly commented 2 years ago

Huh. Very interesting.

On 2021-06-10, when the ESO PV forecast had a particularly large error (ESO over-estimated national PV power by almost 2 GW at around 10:30am), the NWP got the clouds really quite wrong. The NWP predicted almost no cloud over Sheffield and Norwich. But, in reality, there was a bunch of cloud over both locations. And the NWP predicted far too much cloud over the North Sea! This is despite the fact that this plot is showing the NWP at only step 1 (i.e. only 1 hour after the NWP initialisation time).

image

image

image

JackKelly commented 2 years ago

Here's a video showing HRV and NWP cloud cover for the whole day: https://www.youtube.com/watch?v=aZadtsZopsU