Interpolation of weather data

[Stefan]

I have a bit of a vague question.

Weather data is typically provided at an hourly interval. Thus to compute PV power output of our PV plant we take the weather data and solar position as inputs, run it through the model chain, and get our estimated power output for each datapoint in the weather data set. However, within one hour the sun's position can change quite significantly, especially in summer. 

So my question is: does it make sense to (linearly) interpolate the weather data to for example 15-minute, 5-minute or even 1-minute intervals and use that to estimate PV power? Is there any research / are there any papers that explore this? 

[Mark Mikofski]

I don’t advise interpolating or up-sampling weather data to higher frequency, except under very clear sky conditions because during cloudy conditions high frequency irradiance variations could introduce deviations that would not be handled by interpolation. There are papers on synthesizing higher frequency data that  use statistics from a nearby site such as the SURFRAD data sets. Or you can acquire high frequency data sets from providers like NREL PSM-3 and commercial providers like Solcast, CPR, and SolarGIS (full disclosure, Solcast is a DNV owned company)

[cwh...@sandia.gov]

Not an answer to the question you asked, but related: it is common for modeling packages to use solar position at the midpoint of time intervals, in combination with the data at the time point for the interval (let's assume at the interval beginning). The reasoning seems to be to view the single irradiance value as the interval average value, and that the solar position at the midpoint gives a better estimate of the average plane-of-array irradiance. These arguments aren't mathematically exact, since most of the models involves (transposition to plane-of-array, cell temperature, module efficiency) may be nearly, but not exactly, linear in their inputs, so applying the model to the average input doesn't give the average of the output. But the result is better, in some sense, than using the solar position at the beginning of the interval. 

I know I have seen papers quantifying the approximation error of modeling with different time scales, but I don't have a reference at hand other than my own report from some time ago, perhaps its terms can help you find more recent works. https://www.researchgate.net/publication/266557135_Effect_of_Time_Scale_on_Analysis_of_PV_System_Performance

Cheers,

Cliff

[Will Hobbs]

Some additional papers that might be useful are below. They are mostly papers I was involved in, which makes them easier to remember :). 

Most of them are about solar *resource* changing within an hour due to variable clouds, but some of the papers cited within them might be more directly applicable to your specific point about solar *position* changing a lot within an hour. 

• https://doi.org/10.1109/JPHOTOV.2023.3238512
• https://doi.org/10.1063/5.0130265
  
• https://doi.org/10.36227/techrxiv.23837340.v2
  
• https://www.nrel.gov/docs/fy22osti/82812.pdf
  

[Stefan]

Thanks a lot all for responding! It is very appreciated. I didn't expect so many experts here. 

Seems like I have some homework to do so I will go through the resources you have shared and I will let you know if I have any followup questions.

Op dinsdag 5 september 2023 om 22:46:48 UTC+2 schreef will....@gmail.com: