Snow depth as predictor

[sergejlazuk]

Snow Depth should be used as predictor

Detailed Description

Snow depth has large impact and should be considered

Context

Possible Implementation

I think snow depth is available from GFS and ICON?

[peterdudfield]

Hey @sergejlazuk, thanks very much for idea. Yea snow depth is normally available in most NWPs.

We have trained using PV system from the UK, where snow depth plays very little role. However I totally understand that in other parts of the world, it plays an important factor.

There's probably some different options we could go from here.

1. If you have any data to share, we could look at the correlation of snow depth and the power generation?
2. With this data we could even add it to our training and re train the model with this factor
3. We could add a small function at the end of the forecat, thats sets the forecast to 0 if snow depth>0. This seems quite drastic and it could ignore affects like, there's snow on the grown, but its melted already on the PV panels.

[sergejlazuk]

Hi Peter,
Unfortunately I dont have access to any measurements data.
I just noticed this snow depth effect once I was trying to predict PV Generation for German Distribution Grid (here I used realized generation over whole region, not single PV installation) - on some days in Winter my forecast had large error, and after checking open-meteo, it seems those were also the days where there was non-zero snow depth.
But I think setting PV generation to zero is over-simplistic and might underestimate actually produced power.

Did you happen to work with Duett dataset from DWD:
https://www.dwd.de/EN/ourservices/duett_strahlungsmessnetz/ls_duett.html

As Duett accounts for ground measurements (combined those with satellite data), maybe their value incorporate snow effect, but I am not sure.
Duett only has data for previous hour (with ~15Mi delay), but as snow depth impact is persistent, maybe your model could incorporate past Duett values for forecast up to day ahead, just an idea.

Regards,
Sergej

[peterdudfield]

Hey @sergejlazuk

Thanks for this. Yea this makes total sense in Germany, and many other European countries and global countries.

We do work with the DWD ICON dataset. Infact we publish a archive on Hugging Face https://huggingface.co/datasets/openclimatefix/dwd-icon-eu

Yea, thats useful for link. We are using satellite data for our OCF forecasts. For open-source-quartz-solar-forecast we are trying to work out a way to use satellite images. There are some license issues, but happy to explore this more.

[peterdudfield]

@all-contributors please add @sergejlazuk for question?

You ok for this to be added?

[allcontributors]

@peterdudfield

I've put up a pull request to add @sergejlazuk! 🎉