So I’m in Hobart, Tasmania. A beautiful city I might add, with some picturesque upper level cirrus arranged in awesome gravity wave bands. But I’m not here as a cloud tourist, but in fact to present some interesting research I am completing with an undergraduate student (Sonya Wellby) at the Australian Meteorological and Oceanographic Society's Annual Conference (AMOS 2014).
In the course of producing solar forecasts for our ARENA USASEC grant project (I really do need to actually write up a blog post on that at some point), we’ve discovered something interesting about the inclusion of weather data in our machine learning algorithms: it doesn’t seem to help at all.
This is quite strange when you consider that the entire challenge of solar energy prediction is related to the clouds and those are driven by recognizable weather phenomena. So including information like the temperature, wind speed/direction, surface pressure, etc should in theory, help the forecast improve.
But this is not what we’ve found. In our single-site, Support Vector Machine (SVM with linear loss function) model estimates for 10 minute interval data/forecasts, we see an increase in Mean Bias and Root Mean Squared Errors and a decrease in the correlation between predictions/observations:
For hourly data/forecast intervals, we see no improvement in the forecasts with the inclusion of weather data.